TREE BASED MODELING, PREDICTION AND ANALYSIS OF CHAOTIC TIME SERIES

合集下载

Incentive-based modeling and inference of attacker intent, objectives, and strategies

Incentive-Based Modeling and Inference of Attacker Intent,Objectives,and StrategiesPENG LIU and WANYU ZANGPennsylvania State UniversityandMENG YUMonmouth UniversityAlthough the ability to model and infer attacker intent,objectives,and strategies(AIOS)may dramatically advance the literature of risk assessment,harm prediction,and predictive or proactive cyber defense,existing AIOS inference techniques are ad hoc and system or application speciﬁc. In this paper,we present a general incentive-based method to model AIOS and a game-theoretic approach to inferring AIOS.On one hand,we found that the concept of incentives can unify a large variety of attacker intents;the concept of utilities can integrate incentives and costs in such a way that attacker objectives can be practically modeled.On the other hand,we developed a game-theoretic AIOS formalization which can capture the inherent interdependency between AIOS and defender objectives and strategies in such a way that AIOS can be automatically inferred.Finally, we use a speciﬁc case study to show how attack strategies can be inferred in real-world attack–defense scenarios.Categories and Subject Descriptors:C.2.0[Computer-Communication Networks]:Security and ProtectionGeneral Terms:Security,TheoryAdditional Key Words and Phrases:Attacker intent and strategy modeling,attack strategy infer-ence,game theory1.INTRODUCTIONThe ability to model and infer attacker intent,objectives,and strategies(AIOS) may dramatically advance the state of the art of computer security for several reasons.First,for many“very difﬁcult to prevent”attacks such as DDoS,given the speciﬁcation of a system protected by a set of speciﬁc security mechanisms, This work was supported by DARPA and AFRL,AFMC,USAF,under award number F20602-02-1-0216,and by Department of Energy Early Career PI Award.Authors’addresses:P.Liu and W.Zang,School of Information Sciences and Technology, Pennsylvania State University,University Park,PA16802;email:pliu@;M.Yu,De-partment of Computer Science,Monmouth University,West Long Branch,NJ07764. Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proﬁt or direct commercial advantage and that copies show this notice on theﬁrst page or initial screen of a display along with the full citation.Copyrights for components of this work owned by others than ACM must be honored.Abstracting with credit is permitted.To copy otherwise,to republish,to post on servers, to redistribute to lists,or to use any component of this work in other works requires prior speciﬁc permission and/or a fee.Permissions may be requested from Publications Dept.,ACM,Inc.,1515 Broadway,New York,NY10036USA,fax:+1(212)869-0481,or permissions@.C 2005ACM1094-9224/05/0200-0078$5.00ACM Transactions on Information and System Security,Vol.8,No.1,February2005,Pages78–118.Incentive-Based Modeling and Inference of AIOS•79 this ability could tell us which kind of strategies are more likely to be taken by the attacker than the others,even before such an attack happens.Such AIOS inferences may lead to more precise risk assessment and harm prediction.Second,AIOS modeling and inference could be more beneﬁcial during run time.A big security challenge in countering a multiphase,well-planned,care-fully hidden attack from either malicious insiders or outside attackers is“how to make correct proactive(especially predictive)real-time defense decisions during an earlier stage of the attack in such a way that much less harm will be caused without consuming a lot of resources?”Although many proactive de-fense techniques are developed such as sandboxing[Malkhi and Reiter2000] and isolation[Liu et al.2000],making the right proactive defense decisions in real time is very difﬁcult primarily due to the fact that intrusion detection during the early stage of an attack can lead to many false alarms,which could make these proactive defense actions very expensive in terms of both resources and denial of service.Although alert correlation techniques[Cuppens and Miege2002;Ning et al. 2002]may reduce the number of false alarms by correlating a set of alerts into an attack scenario(i.e.,steps involved in an attack)and may even tell which kind of attack actions may follow a given action[Debar and Wespi2001], they are limited in supporting proactive intrusion response in two aspects.(1) When many types of(subsequences of)legitimate actions may follow a given suspicious action,alert correlation can do nothing except for waiting until a more complete attack scenario emerges.However,intrusion response at this moment could be“too late.”(2)When many types of attack actions may follow a given(preparation)action,alert correlation cannot tell which actions are more likely to be taken by the attacker next.As a result,since taking proactive defense actions for each of the attack actions can be too expensive,the response may have to wait until it is clear what attack actions will happen next—perhaps during a later stage of the attack.However,late intrusion response usually means more harm.By contrast,with the ability to model and infer AIOS,given any suspicious action,we can predict the harm that could be caused;then we can make better and affordable proactive intrusion response decisions based on the corresponding risk,the corresponding cost(e.g.,due to the possibility of false alarms),and the attack action inferences.Moreover,the intrusion response time is substantially shortened.However,with a focus on attack characteristics[Landwehr et al.1994]and attack effects[Browne et al.2001;Zou et al.2002],existing AIOS inference tech-niques are ad hoc and system or application speciﬁc[Gordon and Loeb2001; Syverson1997].To systematically model and infer AIOS,we need to distin-guish AIOS from both attack actions and attack effects.Since the same attack action can be issued by two attackers with very different intents and objectives, AIOS cannot be directly inferred from the characteristics of attacks.Although the attacker achieves his or her intents and objectives through attacks and their effects,the mapping from attack actions and/or effects to attacker intents and/or objectives is usually not one-to-one but one-to-many,and more inter-estingly,the(average)cardinality of this mapping can be much larger than the mapping from attacker intents and/or objectives to attack actions and/or ACM Transactions on Information and System Security,Vol.8,No.1,February2005.80•P.Liu et al.effects.This asymmetry nature indicates that in many cases using AIOS mod-els to predict attack actions can be more precise than using the set of actions already taken by the attacker based on either their effects or the causal rela-tionship between them and some other attack actions.1As a result,although a variety of attack taxonomies and attribute databases have been developed, people’s ability to model and infer AIOS,to predict attacks,and to do proactive intrusion response is still very limited.Nevertheless,a good understanding of attacks is the foundation of practical AIOS modeling and inference.In this paper,we present a systematic incentive-based method to model AIOS and a game-theoretic approach to inferring AIOS.On one hand,we found that the concept of incentives can unify a large variety of attacker intents;the con-cept of utilities can integrate incentives and costs in such a way that attacker objectives can be practically modeled.On the other hand,we developed a game-theoretic AIOS formalization which can capture the inherent interdependency between AIOS and defender objectives and strategies in such a way that AIOS can be automatically inferred.Finally,we use a speciﬁc case study to show how attack strategies can be inferred in real-world attack–defense scenarios.The proposed framework,in some sense,is an economics-based framework since it is based on economic incentives,utilities,and payoffs.The rest of the paper is organized as follows.In Section2,we discuss the related work.Section3presents a conceptual,incentive-based framework for AIOS modeling.In Section4,we present a game-theoretic formalization of this framework.Section5addresses show to infer AIOS.In Section6,we use a speciﬁc case study to show how attack strategies can be inferred in real-world attack–defense scenarios.In Section7,we mention several future research issues.2.RELATED WORKThe use of game theory in modeling attackers and defenders has been addressed in several other research.In Syverson[1997],Syverson talks about“good”nodes ﬁghting“evil”nodes in a network and suggests using stochastic games for rea-soning and analysis.In Lye and Wing[2002],Lye and Wing precisely formalize this idea using a general-sum stochastic game model and give a concrete ex-ample in detail where the attacker is attacking a simple enterprise network that provides some Internet services such as web and FTP.A set of speciﬁc states regarding this example are identiﬁed,state-transition probabilities are assumed,and the Nash equilibrium or best-response strategies for the players are computed.In Browne[2000],Browne describes how static games can be used to an-alyze attacks involving complicated and heterogeneous military networks.In his example,a defense team has to defend a network of three hosts against an attacking team’s worms.The defense team can choose either to run a worm 1To illustrate,consider a large space of strategies the attacker may take according to his or her intent and objectives where each strategy is simply a sequence of actions.An attack action may belong to many strategies,and the consequences of the action could satisfy the preconditions of many other actions,but each strategy usually contains only a small number of actions.ACM Transactions on Information and System Security,Vol.8,No.1,February2005.Incentive-Based Modeling and Inference of AIOS•81 detector or not.Depending on the combined attack and defense actions,each outcome has different costs.In Burke[1999],Burke studies the use of repeated games with incomplete information to model attackers and defenders in in-formation warfare.In Hespanha and Bohacek[2001],Hespanha and Bohacek discuss zero-sum routing games where an adversary(or attacker)tries to in-tersect data packets in a computer network.The designer of the network has to ﬁnd routing policies that avoid links that are under the attacker’s surveillance. In Xu and Lee[2003],Xu and Lee use game-theoretical framework to analyze the performance of their proposed DDoS defense system and to guide its design and performance tuning accordingly.Our work is different from the above game theoretic attacker modeling works in several aspects.First,these works focus on speciﬁc attack–defense scenarios, while our work focuses on general AIOS modeling.Second,these works focus on speciﬁc types of game models,for example,static games,repeated games, or stochastic games;while our work focuses on the fundamental characteris-tics of AIOS,and game models are only one possible formalization of our AIOS framework.In addition,our AIOS framework shows the inherent relationship between AIOS and the different types of game models,and identiﬁes the condi-tions under which a speciﬁc type of game models will be feasible and desirable. Third,our work systematically identiﬁes the properties of a good AIOS for-malization.These properties not only can be used to evaluate the merits and limitations of game-theoretic AIOS models,but also can motivate new AIOS models that can improve the above game theory models or even go beyond standard game-theoretic models.In Gordon and Loeb[2001],information security is used as a response to game theoretic competitor analysis systems(CAS)for the purpose of protecting a ﬁrm’s valuable business data from its competitors.Although understanding and predicting the behavior of competitors are key aspects of competitor analysis, the behaviors CAS want to predict are not cyber attacks.Moreover,security is what our game theoretic system wants to model while security is used in Gordon and Loeb[2001]to protect a game-theoretic system.The computational complexity of game-theoretic analysis is investigated in several research.For example,Conitzer and Sandholm[2002]show that both determining whether a pure strategy Bayes–Nash equilibrium exists and de-termining whether a pure strategy Nash equilibrium exists in a stochastic (Markov)game are NP-hard.Moreover,Koller and Milch[2001]show that some speciﬁc knowledge representations,in certain settings,can dramatically speed up equilibriumﬁnding.The marriage of economics and information security has attracted a lot of in-terests recently(a lot of related work can be found at the economics and security resource page maintained by Ross Anderson at /∼rja14 /econsec.html).However,these work focuses on the economics perspective of security(e.g.,security market,security insurance),while our approach is to apply economics concepts to model and infer AIOS.In recent years,it is found that economic mechanism design theory[Clarke 1971;Groves1973;Vickrey1961]can be very valuable in solving a variety of Internet computing problems such as routing,packet scheduling,and web ACM Transactions on Information and System Security,Vol.8,No.1,February2005.82•P.Liu et al.work topology.caching[Feigenbaum et al.2002;Nisan and Ronan2001;Wellman and Walsh2001].Although when market-based mechanisms are used to defend against at-tackers[Wang and Reiter2003],the AIOS are incentive based,which is consis-tent with our framework,market-based computing does not imply an in-depthAIOS model.Finally,it should be noticed that AIOS modeling and inference are very differ-ent from intrusion detection[Lunt1993;McHugh2001;Mukherjee et al.1994].Intrusion detection is based on the characteristics of attacks,while AIOS mod-eling is based on the characteristics of attackers.Intrusion detection focuses onthe attacks that have already happened,while AIOS inference focuses on theattacks that may happen in the future.3.AN INCENTIVE-BASED FRAMEWORK FOR AIOS MODELINGIn this section,we present an incentive-based conceptual model of attackerintent,objectives,and strategies.Our model is quite abstract.To make ourpresentation more tangible,we willﬁrst present the following example,whichwill be used throughout the paper to illustrate our concepts.Example1.In recent years,Internet distributed denial-of-service(DDoS)attacks have increased in frequency,severity,and sophistication and becomea major security threat.When a DDoS attack is launched,a large number ofhosts(called zombies)“controlled”by the attackerﬂood a high volume of pack-ets toward the target(called the victim)to downgrade its service performancesigniﬁcantly or make it unable to deliver any service.In this example,we would model the intent and objectives and infer thestrategies of the attackers that enforce brute-force DDoS attacks.(Althoughsome DDoS attacks with clear signatures,such as SYNﬂooding,can be effec-tively countered,most DDoS attacks without clear signatures,such as brute-force DDoS attacks,are very difﬁcult to defend against since it is not clear whichpackets are DDoS packets and which are not.)An example scenario is shownin Figure1where many zombies(i.e.,a subset of source hosts{S0,...,S64})are ﬂooding a couple of web sites(i.e.,the victims)using normal HTTP requests.Here,Rx.y denotes a router;the bandwidth of each type of links is marked;andthe web sites may stay on different subnets.ACM Transactions on Information and System Security,Vol.8,No.1,February2005.Incentive-Based Modeling and Inference of AIOS•83 Although our modeling and inference framework can handle almost every DDoS defense mechanism,to make this example more tangible,we select pushback[Ioannidis and Bellovin2002],a popular technique,as the security mechanism.Pushback uses aggregates,that is,a collection of packets from one or moreﬂows that have some properties in common,to identify and rate limit the packets that are most likely to cause congestion or DoS.Pushback is a coordinated defense mechanism that typically involves multiple routers.To il-lustrate,consider Figure1again,when router R1.0detects a congestion caused by a set of aggregates,R1.0will not only rate-limit these aggregates,but also request adjacent upstream routers(e.g.,R2.1)to rate-limit the corresponding aggregates via some pushback messages.The effectiveness of pushback can be largely captured by four bandwidth parameters associated with the incoming link to the victims(i.e.,the link that connects R1.0and R0.0):(a)B N,the total bandwidth of this link;(b)B ao,the (amount of)bandwidth occupied by the DoS packets;(c)B lo,the bandwidth occupied by the legitimate packets;(d)B lw,the bandwidth that the legitimate users would occupy if there are no attacks.For example,pushback is effective if after being enforced B ao can become smaller and B lo can become larger.We build our AIOS models on top of the relationships between the attacker and a computer system(i.e.,the defender).In our model,the computer sys-tem can be any kind(e.g.,a network system,a distributed system,a database system).We call it the system for short.For example,in Example1the sys-tem consists of every router on a path from a zombie to a victim.The attacker issues attacks to the system.Each attack is a sequence of attack actions associ-ated with the system.For example,an action can be the sending of a message, the submission of a transaction,the execution of a piece of code,and so on.An attack will cause some effects on the system,that is,transforming the system from one state to another state.For example,in Example1the main attack effects are that many legitimate packets could not reach the victims.Part of the system is a set of speciﬁc security mechanisms.A mechanism can be a piece of software or hardware(e.g.,aﬁrewall,an access controller,an IDS).A mechanism usually involves a sequence of defense actions associated with the system when being activated.For example,in Example1a router sending out a pushback message is a defense action,and this action can trigger the receiving router(s)to take further defense actions.A security mechanism is activated when an event arrives which causes a set of speciﬁc conditions to be satisﬁed.Many of these conditions are associated with the effects of an attack action in reactive defense,or the prediction of an incoming attack action in proactive defense.For example,in Example1a packet arriving at a router is an event.When there is no congestion at the router,this event will not activate any security mechanism.However,when this event leads to“the detection of a congestion”(i.e.,the condition),pushback will be activated.And it is clear that whether this condition can be satisﬁed is dependent upon the accumulated effects of the previous DoS packets arriving at the router.Finally,a defense posture of the system is deﬁned by the set of security mechanisms and the ways they are activated.For example,in Example1,pushback may be conﬁgured ACM Transactions on Information and System Security,Vol.8,No.1,February2005.84•P.Liu et al.to stay at various defense postures based on such parameters as congestion thresholds and target drop rate,which we will explain in Section3.3shortly.The attacker-system relation has several unique characteristics(or proper-ties)that are important in illustrating the principles of our attack strategy inference framework.These properties are as follows.—Intentional Attack Property.Attacks are typically not random.They are planned by the attacker based on some intent and objectives.—Strategy-Interdependency Property.Whether an attack can succeed is depen-dent on how the system is protected.Whether a security mechanism is effec-tive is dependent on how the system is attacked.In other words,the capacity of either an attack or a defense posture should be measured in a relative way.We will deﬁne the notion of strategy shortly.And we will use concrete attack and defense strategies derived from Example1to illustrate this property shortly in Section3.3.—Uncertainty Property.The attacker usually has incomplete information or knowledge about the system,and vice versa.For example,in Example1the attacker usually has uncertainty about how Pushback is conﬁgured when he or she enforces a DDoS attack.3.1Incentive-Based Attacker Intent ModelingDifferent attackers usually have different intents even when they issue the same attack.For example,some attackers attack the system to show off their hacking capacity,some hackers attack the system to remind the administrator of a securityﬂaw,cyber terrorists attack our cyberspace for creating damage, business competitors may attack each other’s information systems to increase their market shares,just to name a few.It is clear that investigating the char-acteristics of each kind of intents involves a lot of effort and complexity,and such complexity actually prevents us from building a general,robust connec-tion between attacker intents and attack actions.This connection is necessary to do almost every kind of attacker behavior inference.We focus on building general yet simple intent models.In particular,we believe that the concept of economic“incentives”can be used to model attacker intent in a general way.In our model,the attacker’s intent is simply to maximize his or her incentives.In other words,the attacker is motivated by the possibility of gaining some incentives.Most,if not all,kinds of intents can be modeled as incentives such as the amount of proﬁt earned,the amount of terror caused, and the amount of satisfaction because of a nice show-off.For an example,in Example1the incentives for the attacker can be the amount of DoS suffered by the legitimate users.For another example,the incentives for an attacker that enforces a worm attack can be the amount of network resources consumed by the worm’s scanning packets plus the amount of DoS caused on certain type of services.We may use economics theory to classify incentives into such categories as money,emotional reward,and fame.To infer attacker intents,we need to be able to compare one incentive with another.Incentives can be compared with each other either qualitatively or ACM Transactions on Information and System Security,Vol.8,No.1,February2005.Incentive-Based Modeling and Inference of AIOS•85 quantitatively.Incentives can be quantiﬁed in several ways.For example,prof-its can be quantiﬁed by such monetary units as dollars.For another exam-ple,in Example1,the attacker’s incentives can be quantiﬁed by two metrics: (a)B ao/B N,which indicates the absolute impact of the DDoS attack;and(b)B lo/B lw,which indicates the relative availability impact of the attack.Accord-ingly,the attacker’s intent is to maximize B ao/B N but minimize B lo/B lw.One critical issue in measuring and comparing incentives is that under different value systems,different comparison results may be obtained.For example,dif-ferent types of people value such incentives as time,fame,and differently.As a result,very misleading attacker strategy inferences could be produced if we use our value system to evaluate the attacker’s incentives.After an attack is enforced,the incentives(e.g.,money,fame)earned by the attacker are dependent on the effects of the attack,which are typically captured by the degradation of a speciﬁc set of security measurements that the system cares about.Each such measurement is associated with a speciﬁc security met-ric.Some widely used categories of security metrics include but not limited to conﬁdentiality,integrity,availability(against denial-of-service),nonrepudia-tion,and authentication.For example,in Example1the major security metrics of the system are(a)B lo,which indicates the absolute availability provided by the system;and(b)B lo/B lw,which indicates the relative availability provided by the system.In our model,we call the set of security metrics that a system wants to protect the metric vector of the system.(Note that different systems may have different metric vectors.)For example,the metric vector for the system in Example1can be simply deﬁned as B lo,B lo/B lw .At time t,the measurements associated with the system’s metric vector are called the security vector of the system at time t,denoted by V s t.As a result,assume an attack starts at time t1 and ends at t2,then the incentives earned by the attacker(via the attack)maybe measured by degradation(V s t1,V s t2),which basically computes the distance be-tween the two security vectors.For example,in Example1assume the securityvector is V s t1= 1000Mbps,100% before the attack and V s t2= 50Mbps,5%after the attack,then degradation(V s t1,V s t2)= −950Mbps,−95% .The above discussion indicates the following property of AIOS inference:—Attack Effect Property.Effects of attacks usually yield more insights about at-tacker intent and objectives than attack actions.For example,in Example1,a DoS packet indicates almost nothing about the attacker’s intent which canonly be seen after some DoS effects are caused.3.2Incentive-Based Attacker Objective ModelingIn real world,many attackers face a set of constraints when issuing an attack, for example,an attacker may have limited resources;a malicious insider may worry about the risk of being arrested and put into jail.However,our intent model assumes no constraints.To model attacker motivations in a more realistic way,we incorporate constraints in our attack objective model.In particular,we classify constraints into two categories:cost constraints and noncost constraints.(a)Cost constraints are constraints on things that the attacker can“buy”or “trade”such as hardware,software,Internet connection,and time.Such things ACM Transactions on Information and System Security,Vol.8,No.1,February2005.86•P.Liu et al.are typically used to measure the cost of an attack.In addition,risk is typically a cost constraint.(b)Noncost constraints are constraints on things that the attacker cannot buy such as religion-based constraints and top secret attacking tools that the attacker may never be able to“buy.”The cost of an attack is not only dependent on the resources needed to en-force the attack,but also dependent on the risk for the attacker to be traced back,arrested,and punished.Based on the relationship between incentives and costs,we classify attackers into two categories:(a)rational attackers have concerns about the costs(and risk)associated with their attacks.That is,when the same incentive can be obtained by two attacks with different costs,ratio-nal attackers will pick the one with a lower cost.(b)Irrational attackers have no concerns about the costs associated with their attacks.They only want to maximize the incentives.Given a set of(cost)constraints,inferring the attack actions of an irrational attacker is not so difﬁcult a task since we need only toﬁnd out“what are the most rewarding attack actions in the eyes of the attacker without violating the constraints?”By contrast,we found that inferring the attack actions of a rational attacker is more challenging.In this paper,we will focus on how to model and infer the IOS of rational attackers.In our model,an attacker’s objective is to maximize his or her utilities through an attack without violating the set of cost and noncost constraints associated with the attacker.The utilities earned by an attacker indicate a distance be-tween the incentives earned by the attacker and the cost of the attack.The dis-tance can be deﬁned in several ways,for example,utilities=incentives−cost, utilities=incentives.Note that the cost of an attack can be measured by a set of cost values which captures both attacking resources and risk.To illustrate,let us revisit Example1.The attacker’s total incentives may be measured byαB ao/B N+(1−α)(1−B lo/B lw),whereαdetermines how the attacker weighs the two aspects of the impact of the DDoS attack.The attack’s costs in this example are not much,though the attacker needs a computer and Internet access to“prepare”the zombies and the needed controls.The cost will become larger when the risk of being traced back is included.Let us assume the cost is a constant numberη.Then the attacker’s utilities can be measured by αB ao/B N+(1−α)(1−B lo/B lw)−η,and the attacker’s objective can be quantiﬁed as MaxαB ao/B N+(1−α)(1−B lo/B lw).3.3Incentive-Based Attacker Strategy ModelingStrategies are taken to achieve objectives.The strategy-interdependency prop-erty indicates that part of a good attacker strategy model should be the defense strategy model because otherwise we will build our AIOS models on top of the assumption that the system never changes its defense posture,which is too restrictive.See that whenever the system’s defense posture is changed,the defense strategy is changed.In our model,attack strategies are deﬁned based on the“battles”between the attacker and the system.Each attack triggers a battle which usually involves multiple phases.(For example,many worm-based attacks involve such phases ACM Transactions on Information and System Security,Vol.8,No.1,February2005.。

H2O.ai 自动化机器学习蓝图：人类中心化、低风险的 AutoML 框架说明书

Beyond Reason CodesA Blueprint for Human-Centered,Low-Risk AutoML H2O.ai Machine Learning Interpretability TeamH2O.aiMarch21,2019ContentsBlueprintEDABenchmarkTrainingPost-Hoc AnalysisReviewDeploymentAppealIterateQuestionsBlueprintThis mid-level technical document provides a basic blueprint for combining the best of AutoML,regulation-compliant predictive modeling,and machine learning research in the sub-disciplines of fairness,interpretable models,post-hoc explanations,privacy and security to create a low-risk,human-centered machine learning framework.Look for compliance mode in Driverless AI soon.∗Guidance from leading researchers and practitioners.Blueprint†EDA and Data VisualizationKnow thy data.Automation implemented inDriverless AI as AutoViz.OSS:H2O-3AggregatorReferences:Visualizing Big DataOutliers through DistributedAggregation;The Grammar ofGraphicsEstablish BenchmarksEstablishing a benchmark from which to gauge improvements in accuracy,fairness, interpretability or privacy is crucial for good(“data”)science and for compliance.Manual,Private,Sparse or Straightforward Feature EngineeringAutomation implemented inDriverless AI as high-interpretabilitytransformers.OSS:Pandas Proﬁler,Feature ToolsReferences:Deep Feature Synthesis:Towards Automating Data ScienceEndeavors;Label,Segment,Featurize:A Cross Domain Framework forPrediction EngineeringPreprocessing for Fairness,Privacy or SecurityOSS:IBM AI360References:Data PreprocessingTechniques for Classiﬁcation WithoutDiscrimination;Certifying andRemoving Disparate Impact;Optimized Pre-processing forDiscrimination Prevention;Privacy-Preserving Data MiningRoadmap items for H2O.ai MLI.Constrained,Fair,Interpretable,Private or Simple ModelsAutomation implemented inDriverless AI as GLM,RuleFit,Monotonic GBM.References:Locally InterpretableModels and Eﬀects Based onSupervised Partitioning(LIME-SUP);Explainable Neural Networks Based onAdditive Index Models(XNN);Scalable Bayesian Rule Lists(SBRL)LIME-SUP,SBRL,XNN areroadmap items for H2O.ai MLI.Traditional Model Assessment and DiagnosticsResidual analysis,Q-Q plots,AUC andlift curves conﬁrm model is accurateand meets assumption criteria.Implemented as model diagnostics inDriverless AI.Post-hoc ExplanationsLIME,Tree SHAP implemented inDriverless AI.OSS:lime,shapReferences:Why Should I Trust You?:Explaining the Predictions of AnyClassiﬁer;A Uniﬁed Approach toInterpreting Model Predictions;PleaseStop Explaining Black Box Models forHigh Stakes Decisions(criticism)Tree SHAP is roadmap for H2O-3;Explanations for unstructured data areroadmap for H2O.ai MLI.Interlude:The Time–Tested Shapley Value1.In the beginning:A Value for N-Person Games,19532.Nobel-worthy contributions:The Shapley Value:Essays in Honor of Lloyd S.Shapley,19883.Shapley regression:Analysis of Regression in Game Theory Approach,20014.First reference in ML?Fair Attribution of Functional Contribution in Artiﬁcialand Biological Networks,20045.Into the ML research mainstream,i.e.JMLR:An Eﬃcient Explanation ofIndividual Classiﬁcations Using Game Theory,20106.Into the real-world data mining workﬂow...ﬁnally:Consistent IndividualizedFeature Attribution for Tree Ensembles,20177.Uniﬁcation:A Uniﬁed Approach to Interpreting Model Predictions,2017Model Debugging for Accuracy,Privacy or SecurityEliminating errors in model predictions bytesting:adversarial examples,explanation ofresiduals,random attacks and“what-if”analysis.OSS:cleverhans,pdpbox,what-if toolReferences:Modeltracker:RedesigningPerformance Analysis Tools for MachineLearning;A Marauder’s Map of Security andPrivacy in Machine Learning:An overview ofcurrent and future research directions formaking machine learning secure and privateAdversarial examples,explanation ofresiduals,measures of epistemic uncertainty,“what-if”analysis are roadmap items inH2O.ai MLI.Post-hoc Disparate Impact Assessment and RemediationDisparate impact analysis can beperformed manually using Driverless AIor H2O-3.OSS:aequitas,IBM AI360,themisReferences:Equality of Opportunity inSupervised Learning;Certifying andRemoving Disparate ImpactDisparate impact analysis andremediation are roadmap items forH2O.ai MLI.Human Review and DocumentationAutomation implemented as AutoDocin Driverless AI.Various fairness,interpretabilityand model debugging roadmapitems to be added to AutoDoc.Documentation of consideredalternative approaches typicallynecessary for compliance.Deployment,Management and MonitoringMonitor models for accuracy,disparateimpact,privacy violations or securityvulnerabilities in real-time;track modeland data lineage.OSS:mlﬂow,modeldb,awesome-machine-learning-opsmetalistReference:Model DB:A System forMachine Learning Model ManagementBroader roadmap item for H2O.ai.Human AppealVery important,may require custom implementation for each deployment environment?Iterate:Use Gained Knowledge to Improve Accuracy,Fairness, Interpretability,Privacy or SecurityImprovements,KPIs should not be restricted to accuracy alone.Open Conceptual QuestionsHow much automation is appropriate,100%?How to automate learning by iteration,reinforcement learning?How to implement human appeals,is it productizable?ReferencesThis presentation:https:///navdeep-G/gtc-2019/blob/master/main.pdfDriverless AI API Interpretability Technique Examples:https:///h2oai/driverlessai-tutorials/tree/master/interpretable_ml In-Depth Open Source Interpretability Technique Examples:https:///jphall663/interpretable_machine_learning_with_python https:///navdeep-G/interpretable-ml"Awesome"Machine Learning Interpretability Resource List:https:///jphall663/awesome-machine-learning-interpretabilityAgrawal,Rakesh and Ramakrishnan Srikant(2000).“Privacy-Preserving Data Mining.”In:ACM Sigmod Record.Vol.29.2.URL:/cs/projects/iis/hdb/Publications/papers/sigmod00_privacy.pdf.ACM,pp.439–450.Amershi,Saleema et al.(2015).“Modeltracker:Redesigning Performance Analysis Tools for Machine Learning.”In:Proceedings of the33rd Annual ACM Conference on Human Factors in Computing Systems.URL: https:///en-us/research/wp-content/uploads/2016/02/amershi.CHI2015.ModelTracker.pdf.ACM,pp.337–346.Calmon,Flavio et al.(2017).“Optimized Pre-processing for Discrimination Prevention.”In:Advances in Neural Information Processing Systems.URL:/paper/6988-optimized-pre-processing-for-discrimination-prevention.pdf,pp.3992–4001.Feldman,Michael et al.(2015).“Certifying and Removing Disparate Impact.”In:Proceedings of the21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.URL:https:///pdf/1412.3756.pdf.ACM,pp.259–268.Hardt,Moritz,Eric Price,Nati Srebro,et al.(2016).“Equality of Opportunity in Supervised Learning.”In: Advances in neural information processing systems.URL:/paper/6374-equality-of-opportunity-in-supervised-learning.pdf,pp.3315–3323.Hu,Linwei et al.(2018).“Locally Interpretable Models and Eﬀects Based on Supervised Partitioning (LIME-SUP).”In:arXiv preprint arXiv:1806.00663.URL:https:///ftp/arxiv/papers/1806/1806.00663.pdf.Kamiran,Faisal and Toon Calders(2012).“Data Preprocessing Techniques for Classiﬁcation Without Discrimination.”In:Knowledge and Information Systems33.1.URL:https:///content/pdf/10.1007/s10115-011-0463-8.pdf,pp.1–33.Kanter,James Max,Owen Gillespie,and Kalyan Veeramachaneni(2016).“Label,Segment,Featurize:A Cross Domain Framework for Prediction Engineering.”In:Data Science and Advanced Analytics(DSAA),2016 IEEE International Conference on.URL:/static/papers/DSAA_LSF_2016.pdf.IEEE,pp.430–439.Kanter,James Max and Kalyan Veeramachaneni(2015).“Deep Feature Synthesis:Towards Automating Data Science Endeavors.”In:Data Science and Advanced Analytics(DSAA),2015.366782015.IEEEInternational Conference on.URL:https:///EVO-DesignOpt/groupWebSite/uploads/Site/DSAA_DSM_2015.pdf.IEEE,pp.1–10.Keinan,Alon et al.(2004).“Fair Attribution of Functional Contribution in Artiﬁcial and Biological Networks.”In:Neural Computation16.9.URL:https:///profile/Isaac_Meilijson/publication/2474580_Fair_Attribution_of_Functional_Contribution_in_Artificial_and_Biological_Networks/links/09e415146df8289373000000/Fair-Attribution-of-Functional-Contribution-in-Artificial-and-Biological-Networks.pdf,pp.1887–1915.Kononenko,Igor et al.(2010).“An Eﬃcient Explanation of Individual Classiﬁcations Using Game Theory.”In: Journal of Machine Learning Research11.Jan.URL:/papers/volume11/strumbelj10a/strumbelj10a.pdf,pp.1–18.Lipovetsky,Stan and Michael Conklin(2001).“Analysis of Regression in Game Theory Approach.”In:Applied Stochastic Models in Business and Industry17.4,pp.319–330.Lundberg,Scott M.,Gabriel G.Erion,and Su-In Lee(2017).“Consistent Individualized Feature Attribution for Tree Ensembles.”In:Proceedings of the2017ICML Workshop on Human Interpretability in Machine Learning(WHI2017).Ed.by Been Kim et al.URL:https:///pdf?id=ByTKSo-m-.ICML WHI2017,pp.15–21.Lundberg,Scott M and Su-In Lee(2017).“A Uniﬁed Approach to Interpreting Model Predictions.”In: Advances in Neural Information Processing Systems30.Ed.by I.Guyon et al.URL:/paper/7062-a-unified-approach-to-interpreting-model-predictions.pdf.Curran Associates,Inc.,pp.4765–4774.Papernot,Nicolas(2018).“A Marauder’s Map of Security and Privacy in Machine Learning:An overview of current and future research directions for making machine learning secure and private.”In:Proceedings of the11th ACM Workshop on Artiﬁcial Intelligence and Security.URL:https:///pdf/1811.01134.pdf.ACM.Ribeiro,Marco Tulio,Sameer Singh,and Carlos Guestrin(2016).“Why Should I Trust You?:Explaining the Predictions of Any Classiﬁer.”In:Proceedings of the22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.URL:/kdd2016/papers/files/rfp0573-ribeiroA.pdf.ACM,pp.1135–1144.Rudin,Cynthia(2018).“Please Stop Explaining Black Box Models for High Stakes Decisions.”In:arXiv preprint arXiv:1811.10154.URL:https:///pdf/1811.10154.pdf.Shapley,Lloyd S(1953).“A Value for N-Person Games.”In:Contributions to the Theory of Games2.28.URL: http://www.library.fa.ru/files/Roth2.pdf#page=39,pp.307–317.Shapley,Lloyd S,Alvin E Roth,et al.(1988).The Shapley Value:Essays in Honor of Lloyd S.Shapley.URL: http://www.library.fa.ru/files/Roth2.pdf.Cambridge University Press.Vartak,Manasi et al.(2016).“Model DB:A System for Machine Learning Model Management.”In: Proceedings of the Workshop on Human-In-the-Loop Data Analytics.URL:https:///~matei/papers/2016/hilda_modeldb.pdf.ACM,p.14.Vaughan,Joel et al.(2018).“Explainable Neural Networks Based on Additive Index Models.”In:arXiv preprint arXiv:1806.01933.URL:https:///pdf/1806.01933.pdf.Wilkinson,Leland(2006).The Grammar of Graphics.—(2018).“Visualizing Big Data Outliers through Distributed Aggregation.”In:IEEE Transactions on Visualization&Computer Graphics.URL:https:///~wilkinson/Publications/outliers.pdf.Yang,Hongyu,Cynthia Rudin,and Margo Seltzer(2017).“Scalable Bayesian Rule Lists.”In:Proceedings of the34th International Conference on Machine Learning(ICML).URL:https:///pdf/1602.08610.pdf.。

pride 内存数据库使用说明

PRIDE:A Data Abstraction Layer for Large-Scale2-tier Sensor NetworksWoochul Kang University of Virginia Email:wk5f@Sang H.SonUniversity of VirginiaEmail:son@John A.StankovicUniversity of VirginiaEmail:stankovic@Abstract—It is a challenging task to provide timely access to global data from sensors in large-scale sensor network applica-tions.Current data storage architectures for sensor networks have to make trade-offs between timeliness and scalability. PRIDE is a data abstraction layer for2-tier sensor networks, which enables timely access to global data from the sensor tier to all participating nodes in the upper storage tier.The design of PRIDE is heavily inﬂuenced by collaborative real-time ap-plications such as search-and-rescue tasks for high-rise building ﬁres,in which multiple devices have to collect and manage data streams from massive sensors in cooperation.PRIDE achieves scalability,timeliness,andﬂexibility simultaneously for such applications by combining a model-driven full replication scheme and adaptive data quality control mechanism in the storage-tier. We show the viability of the proposed solution by implementing and evaluating it on a large-scale2-tier sensor network testbed. The experiment results show that the model-driven replication provides the beneﬁt of full replication in a scalable and controlled manner.I.I NTRODUCTIONRecent advances in sensor technology and wireless connec-tivity have paved the way for next generation real-time appli-cations that are highly data-driven,where data represent real-world status.For many of these applications,data streams from sensors are managed and processed by application-speciﬁc devices such as PDAs,base stations,and micro servers.Fur-ther,as sensors are deployed in increasing numbers,a single device cannot handle all sensor streams due to their scale and geographic distribution.Often,a group of such devices need to collaborate to achieve a common goal.For instance,during a search-and-rescue task for a buildingﬁre,while PDAs carried byﬁreﬁghters collect data from nearby sensors to check the dynamic status of the building,a team of suchﬁreﬁghters have to collaborate by sharing their locally collected real-time data with peerﬁreﬁghters since each individualﬁreﬁghter has only limited information from nearby sensors[1].The building-wide situation assessment requires fusioning data from all(or most of)ﬁreﬁghters.As this scenario shows,lots of future real-time applications will interact with physical world via large numbers of un-derlying sensors.The data from the sensors will be managed by distributed devices in cooperation.These devices can be either stationary(e.g.,base stations)or mobile(e.g.,PDAs and smartphones).Sharing data,and allowing timely access to global data for each participating entity is mandatory for suc-cessful collaboration in such distributed real-time applications.Data replication[2]has been a key technique that enables each participating entity to share data and obtain an understanding of the global status without the need for a central server. In particular,for distributed real-time applications,the data replication is essential to avoid unpredictable communication delays[3][4].PRIDE(Predictive Replication In Distributed Embedded systems)is a data abstraction layer for devices performing collaborative real-time tasks.It is linked to an application(s) at each device,and provides transparent and timely access to global data from underlying sensors via a scalable and robust replication mechanism.Each participating device can transparently access the global data from all underlying sen-sors without noticing whether it is from local sensors,or from remote sensors,which are covered by peer devices. Since global data from all underlying sensors are available at each device,queries on global spatio-temporal data can be efﬁciently answered using local data access methods,e.g.,B+ tree indexing,without further communication.Further,since all participating devices share the same set of data,any of them can be a primary device that manages a sensor.For example,when entities(either sensor nodes or devices)are mobile,any device that is close to a sensor node can be a primary storage node of the sensor node.Thisﬂexibility via decoupling the data source tier(sensors)from the storage tier is very important if we consider the highly dynamic nature of wireless sensor network applications.Even with these advantages,the high overhead of repli-cation limits its applicability[2].Since potentially a vast number of sensor streams are involved,it is not generally possible to propagate every sensor measurement to all devices in the system.Moreover,the data arrival rate can be high and unpredictable.During critical situations,the data rates can signiﬁcantly increase and exceed system capacity.If no corrective action is taken,queues will form and the laten-cies of queries will increase without bound.In the context of centralized systems,several intelligent resource allocation schemes have been proposed to dynamically control the high and unpredictable rate of sensor streams[5][6][7].However, no work has been done in the context of distributed and replicated systems.In this paper,we focus on providing a scalable and robust replication mechanism.The contributions of this paper are: 1)a model-driven scalable replication mechanism,which2signiﬁcantly reduces the overall communication and computation overheads,2)a global snapshot management scheme for efﬁcientsupport of spatial queries on global data,3)a control-theoretic quality-of-data management algo-rithm for robustness against unpredictable workload changes,and4)the implementation and evaluation of the proposed ap-proach on a real device with realistic workloads.To make the replication scalable,PRIDE provides a model-driven replication scheme,in which the models of sensor streams are replicated to peer storage nodes,instead of data themselves.Once a model for a sensor stream is replicated from a primary storage node of the sensor to peer nodes,the updates from the sensor are propagated to peer nodes only if the prediction from the current model is not accurate enough. Our evaluation in Section5shows that this model-driven approach makes PRIDE highly scalable by signiﬁcantly re-ducing the communication/computation overheads.Moreover, the Kalmanﬁlter-based modeling technique in PRIDE is light-weight and highly adaptable because it dynamically adjusts its model parameters at run-time without training.Spatial queries on global data are efﬁciently supported by taking snapshots from the models periodically.The snapshot is an up-to-date reﬂection of the monitored situation.Given this fresh snapshot,PRIDE supports a rich set of local data orga-nization mechanisms such as B+tree indexing to efﬁciently process spatial queries.In PRIDE,the robustness against unpredictable workloads is achieved by dynamically adjusting the precision bounds at each node to maintain a proper level of system load,CPU utilization in particular.The coordination is made among the nodes such that relatively under-loaded nodes synchronize their precision bound with an relatively overloaded node. Using this coordination,we ensure that the congestion at the overloaded node is effectively resolved.To show the viability of the proposed approach,we imple-mented a prototype of PRIDE on a large-scale testbed com-posed of Nokia N810Internet tablets[8],a cluster computer, and a realistic sensor stream generator.We chose Nokia N810 since it represents emerging ubiquitous computing platforms such as PDAs,smartphones,and mobile computers,which will be expected to interact with ubiquitous sensors in the near future.Based on the prototype implementation,we in-vestigated system performance attributes such as communica-tion/computation loads,energy efﬁciency,and robustness.Our evaluation results demonstrate that PRIDE takes advantage of full replication in an efﬁcient,highly robust and scalable manner.The rest of this paper is organized as follows.Section2 presents the overview of PRIDE.Section3presents the details of the model-driven replication.Section4discusses our pro-totype implemention,and Section5presents our experimental results.We present related work in Section6and conclusions in Section7.II.O VERVIEW OF PRIDEA.System ModelFig.1.A collaborative application on a2-tier sensor network. PRIDE envisions2-tier sensor network systems with a sensor tier and a storage tier as shown in Figure1.The sensor tier consists of a large number of cheap and simple sensors;S={s1,s2,...,s n},where s i is a sensor.Sensors are assumed to be highly constrained in resources,and per-form only primitive functions such as sensing and multi-hop communication without local storage.Sensors stream data or events to a nearest storage node.These sensors can be either stationary or mobile;e.g.,sensors attached to aﬁreﬁghter are mobile.The storage tier consists of more powerful devices such as PDAs,smartphones,and base stations;D={d1,d2,...,d m}, where d i is a storage node.These devices are relatively resource-rich compared with sensor nodes.However,these devices also have limited resources in terms of processor cycles,memory,power,and bandwidth.Each storage node provides in-network storage for underlying sensors,and stores data from sensors in its vicinity.Each node supports multiple radios;an802.11radio to connect to a wireless mesh network and a802.15.4to communicate with underlying sensors.Each node in this tier can be either stationary(e.g.,base stations), or mobile(e.g.,smartphones and PDAs).The sensor tier and the storage tier have loose coupling; the storage node,which a sensor belongs to,can be changed dynamically without coordination between the two tiers.This loose coupling is required in many sensor network applications if we consider the highly dynamic nature of such systems.For example,the mobility of sensors and storage nodes makes the system design very complex and inﬂexible if two tiers are tightly coupled;a complex group management and hand-off procedure is required to handle the mobility of entities[9]. Applications at each storage node are linked to the PRIDE layer.Applications issue queries to underlying PRIDE layer either autonomously,or by simply forwarding queries from external users.In the search-and-rescue task example,each storage node serves as both an in-network data storage for nearby sensors and a device to run autonomous real-time applications for the mission;the applications collect data by issuing queries and analyzing the situation to report results to theﬁreﬁghter.Afterwards,a node refers to a storage node if it is not explicitly stated.3Fig.2.The architecture of PRIDE(Gray boxes).age ModelIn PRIDE,all nodes in the storage tier are homogeneous in terms of their roles;no asymmetrical function is placed on a sub-group of the nodes.All or part of the nodes in the storage tier form a replication group R to share the data from underlying sensors,where R⊂D.Once a node joins the replication group,updates from its local sensors are propagated to peer nodes;conversely,the node can receive updates from remote sensors via peer nodes.Any storage node,which is receiving updates directly from a sensor,becomes a primary node for the sensor,and it broadcasts the updates from the sensor to peer nodes.However,it should be noted that,as will be shown in Section3,the PRIDE layer at each node performs model-driven replication,instead of replicating sensor data,to make the replication efﬁcient and scalable.PRIDE is characterized by the queries that it supports. PRIDE supports both temporal queries on each individual sensor stream and spatial queries on current global data.Tem-poral queries on sensor s i’s historical data can be answered using the model for s i.An example of temporal query is “What is the value of sensor s i5minutes ago?”For spatial queries,each storage node provides a snapshot on the entire set of underlying sensors(both local and remote sensors.)The snapshot is similar to a view in database ing the snapshot,PRIDE provides traditional data organization and access methods for efﬁcient spatial query processing.The access methods can be applied to any attributes,e.g.,sensor value,sensor ID,and location;therefore,value-based queries can be efﬁciently supported.Basic operations on the access methods such as insertion,deletion,retrieval,and the iterating cursors are supported.Special operations such as join cursors for join operations are also supported by making indexes to multiple attributes,e.g.,temperature and location attributes. This join operation is required to efﬁciently support complex spatial queries such as“Return the current temperatures of sensors located at room#4.”III.PRIDE D ATA A BSTRACTION L AYERThe architecture of PRIDE is shown in Figure2.PRIDE consists of three key components:(i)ﬁlter&prediction engine,which is responsible for sensor streamﬁltering,model update,and broadcasting of updates to peer nodes,(ii)query processor,which handles queries on spatial and temporal data by using a snapshot and temporal models,respectively,and (iii)feedback controller,which determines proper precision bounds of data for scalability and overload protection.A.Filter&Prediction EngineThe goals ofﬁlter&prediction engine are toﬁlter out updates from local sensors using models,and to synchronize models at each storage node.The premise of using models is that the physical phenomena observed by sensors can be captured by models and a large amount of sensor data can be ﬁltered out using the models.In PRIDE,when a sensor Input:update v from sensor s iˆv=prediction from model for s i;1if|ˆv−v|≥δthen2broadcast to peer storage nodes;3update data for s i in the snapshot;4update model m i for s i;5store to cache for later temporal query processing;6else7discard v(or store for logging);8end9Algorithm2:OnUpdateFromPeer.stream s i is covered by PRIDE replication group R,each storage node in R maintains a model m i for s i.Therefore, all storage nodes in R maintain a same set of synchronized models,M={m1,m2,...,m n},for all sensor streams in underlying sensor tier.Each model m i for sensor s i are synchronized at run-time by s i’s current primary storage node (note that s i’s primary node can change during run-time because of the network topology changes either at sensor tier or storage tier).Algorithms1and2show the basic framework for model synchronization at a primary node and peer nodes,respec-tively.In Algorithm1,when an update v is received from sensor s i to its primary storage node d j,the model m i is looked up,and a prediction is made using m i.If the gap between the predicted value from the model,ˆv,and the sensor update v is less than the precision boundδ(line2),then the new data is discarded(or saved locally for logging.)This implies that the current models(both at the primary node and the peer nodes)are precise enough to predict the sensor output with the given precision bound.However,if the gap is bigger than the precision bound,this implies that the model cannot capture the current behavior of the sensor output.In this case, m i at the primary node is updated and v is broadcasted to all peer nodes(line3).In Algorithm2,as a reaction to the broadcast from d j,each peer node receives a new update v and updates its own model m i with v.The value v is stored in local caches at all nodes for later temporal query processing.4As shown in the Algorithms,the communication among nodes happens only when the model is not precise enough. Models,Filtering,and Prediction So far,we have not discussed a speciﬁc modeling technique in PRIDE.Several distinctive requirements guide the choice of modeling tech-nique in PRIDE.First,the computation and communication costs for model maintenance should be low since PRIDE han-dles a large number of sensors(and corresponding models for each sensor)with collaboration of multiple nodes.The cost of model maintenance linearly increases to the number of sensors. Second,the parameters of models should be obtained without an extensive learning process,because many collaborative real-time applications,e.g.,a search-and-rescue task in a building ﬁre,are short-term and deployed without previous monitoring history.A statistical model that needs extensive historical data for model training is less applicable even with their highly efﬁcientﬁltering and prediction performance.Finally, the modeling should be general enough to be applied to a broad range of applications.Ad-hoc modeling techniques for a particular application cannot be generally used for other applications.Since PRIDE is a data abstraction layer for wide range of collaborative applications,the generality of modeling is important.To this end,we choose to use Kalmanﬁlter [10][6],which provides a systematic mechanism to estimate past,current,and future state of a system from noisy measure-ments.A short summary on Kalmanﬁlter follows.Kalman Filter:The Kalmanﬁlter model assumes the true state at time k is evolved from the state at(k−1)according tox k=F k x k−1+w k;(1) whereF k is the state transition matrix relating x k−1to x k;w k is the process noise,which follows N(0,Q k);At time k an observation z k of the true state x k is made according toz k=H k x k+v k(2) whereH k is the observation model;v k is the measurement noise,which follows N(0,R k); The Kalmanﬁlter is a recursive minimum mean-square error estimator.This means that only the estimated state from the previous time step and the current measurement are needed to compute the estimate for the current and future state. In contrast to batch estimation techniques,no history of observations is required.In what follows,the notationˆx n|m represents the estimate of x at time n given observations up to,and including time m.The state of aﬁlter is deﬁned by two variables:ˆx k|k:the estimate of the state at time k givenobservations up to time k.P k|k:the error covariance matrix(a measure of theestimated accuracy of the state estimate). Kalmanﬁlter has two distinct phases:Predict and Update. The predict phase uses the state estimate from the previous timestep k−1to produce an estimate of the state at the next timestep k.In the update phase,measurement information at the current timestep k is used to reﬁne this prediction to arrive at a new more accurate state estimate,again for the current timestep k.When a new measurement z k is available from a sensor,the true state of the sensor is estimates using the previous predictionˆx k|k−1,and the weighted prediction error. The weight is called Kalman gain K k,and it is updated on each prediction/update cycle.The true state of the sensor is estimated as follows,ˆx k|k=ˆx k|k−1+K k(z k−H kˆx k|k−1).(3)P k|k=(I−K k H k)P k|k−1.(4) The Kalman gain K k is updated as follows,K k|k=P k|k−1H T k(H k P k|k−1H T k+R k).(5) At each prediction step,the next state of the sensor is predicted by,ˆx k|k−1=F kˆx k−1|k−1.(6) Example:For instance,a temperature sensor can be described by the linear state space,x k= x dxdtis the derivative of the temperature with respect to time.As a new(noisy)measurement z k arrives from the sensor1,the true state and model parameters are estimated by Equations3-5.The future state of the sensor at(k+1)th time step after∆t can be predicted using the Equation6, where the state transition matrix isF= 1∆t01 .(7) It should be noted that the parameters for Kalmanﬁlter,e.g., K and P,do not have to be accurate in the beginning;they can be estimated at run-time and their accuracy improves gradually by having more sensor measurements.We do not need massive past data for modeling at deployment time.In addition,the update cycle of Kalmanﬁlter(Equations3-5) is performed at all storage nodes when a new measurement is broadcasted as shown in Algorithm1(line5)and Algorithm2 (line2).No further communication is required to synchronize the parameters of the models.Finally,as will be shown in Section5,the prediction/update cycle of Kalmanﬁlter incurs insigniﬁcant overhead to the system.1Note that the temperature component of zk is directly acquired from the sensor,and dx5B.Query ProcessorThe query processor of PRIDE supports both temporal queries and spatial queries with planned extension to support spatio-temporal queries.Temporal Queries:Historical data for each sensor stream can be processed in any storage node by exploiting data at the local cache and linear smoother[10].Unlike the estimation of current and future states using one Kalmanﬁlter,the optimized estimation of historical data(sometimes called smoothing) requires two Kalmanﬁlters,a forwardﬁlterˆx and a backward ﬁlterˆx b.Smoothing is a non-real-time data processing scheme that uses all measurements between0and T to estimate the state of a system at a certain time t,where0≤t≤T(see Figure3.)The smoothed estimateˆx(t|T)can be obtained as a linear combination of the twoﬁlters as follows.ˆx(t|T)=Aˆx(t)+A′ˆx(t)b,(8) where A and A′are weighting matrices.For detailed discus-sion on smoothing techniques using Kalmanﬁlters,the reader is referred to[10].Fig.3.Smoothing for temporal query processing.Spatial Queries:Each storage node maintains a snapshot for all underlying local and remote sensors to handle queries on global spatial data.Each element(or data object)of the snapshot is an up-to-date value from the corresponding sensor.The snapshot is dynamically updated either by new measurements from sensors or by models2.The Algorithm1 (line4)and Algorithm2(line1)show the snapshot updates when a new observation is pushed from a local sensor and a peer node,respectively.As explained in the previous section, there is no communication among storage nodes when models well represent the current observations from sensors.When there is no update from peer nodes,the freshness of values in the snapshot deteriorate over time.To maintain the freshness of the snapshot even when there is no updates from peer nodes,each value in the snapshot is periodically updated by its local model.Each storage node can estimate the current state of sensor s i using Equation6without communication to the primary storage node of s i.For example,a temperature after30seconds can be predicted by setting∆t of transition matrix in Equation7to30seconds.The period of update of data object i for sensor s i is determined,such that the precision boundδis observed. Intuitively,when a sensor value changes rapidly,the data object should be updated more frequently to make the data object in the snapshot valid.In the example of Section3.1.1, 2Note that the data structures for the snapshot such as indexes are also updated when each value of the snapshot is updated.the period can be dynamically estimated as follows:p[i]=δ/dxdtis the absolute validity interval(avi)before the data object in the snapshot violates the precision bound,which is±δ.The update period should be as short as the half of the avi to make the data object fresh[11].Since each storage node has an up-to-date snapshot,spatial queries on global data from sensors can be efﬁciently han-dled using local data access methods(e.g.,B+tree)without incurring further communication delays.(a)δ=5C(b)δ=10CFig.4.Varying data precision.Figure4shows how the value of one data object in the snapshot changes over time when we apply different precision bounds.As the precision bound is getting bigger,the gap be-tween the real state of the sensor(dashed lines)and the current value at the snapshot(solid lines)increases.In the solid lines, the discontinued points are where the model prediction and the real measurement from the sensor are bigger than the precision bound,and subsequent communication is made among storage nodes for model synchronization.For applications and users, maintaining the smaller precision bound implies having a more accurate view on the monitored situation.However, the overhead also increases as we have the smaller precision bound.Given the unpredictable data arrival rates and resource constraints,compromising the data quality for system sur-vivability is unavoidable in many situations.In PRIDE,we consider processor cycles as the primary limited resource,and the resource allocation is performed to maintain the desired CPU utilization.The utilization control is used to enforce appropriate schedulable utilization bounds of applications can be guaranteed despite signiﬁcant uncertainties in system work-loads[12][5].In utilization control,it is assumed that any cycles that are recovered as a result of control in PRIDE layer are used sensibly by the scheduler in the application layer to relieve the congestion,or to save power[12][5].It can also enhance system survivability by providing overload protection against workloadﬂuctuation.Speciﬁcation:At each node,the system speciﬁcation U,δmax consists of a utilization speciﬁcation U and the precision speciﬁcationδmax.The desired utilization U∈[0..1]gives the required CPU utilization not to overload the system while satisfying the target system performance6 such as latency,and energy consumption.The precisionspeciﬁcationδmax denotes the maximum tolerable precision bound.Note there is no lower bound on the precision as in general users require a precision bound as short as possible (if the system is not overloaded.)Local Feedback Control to Guarantee the System Spec-iﬁcation:Using feedback control has shown to be very effec-tive for a large class of computing systems that exhibit unpre-dictable workloads and model inaccuracies[13].Therefore,to guarantee the system speciﬁcation without a priori knowledge of the workload or accurate system model we apply feedbackcontrol.Fig.5.The feedback control loop.The overall feedback control loop at each storage node is shown in Figure5.Let T is the sampling period.The utilization u(k)is measured at each sampling instant0T,1T,2T,...and the difference between the target utilization and u(k)is fed into the ing the difference,the controller computes a local precision boundδ(k)such that u(k)converges to U. Theﬁrst step for local controller design is modeling the target system(storage node)by relatingδ(k)to u(k).We model the the relationship betwenδ(k)and u(k)by using proﬁling and statistical methods[13].Sinceδ(k)has higher impact on u(k)as the size of the replication group increases, we need different models for different sizes of the group. We change the number of members of the replication group exponentially from2to64and have tuned a set ofﬁrst order models G n(z),where n∈{2,4,8,16,32,64}.G n(z)is the z-transform transfer function of theﬁrst-order models,in which n is the size of the replication group.After the modeling, we design a controller for the model.We have found that a proportional integral(PI)controller[13]is sufﬁcient in terms of providing a zero steady-state error,i.e.,a zero difference between u(k)and the target utilization bound.Further,a gain scheduling technique[13]have been used to apply different controller gains for different size of replication groups.For instance,the gain for G32(z)is applied if the size of a replication group is bigger than24and less than or equal to48. Due to space limitation we do not provide a full description of the design and tuning methods.Coordination among Replication Group Members:If each node independently sets its own precision bound,the net precision bound of data becomes unpredictable.For example, at node d j,the precision bounds for local sensor streams are determined by d j itself while the precision bounds for remote sensor streams are determined by their own primary storage nodes.PRIDE takes a conservative approach in coordinating stor-age nodes in the group.As Algorithm3shows,the global precision bound for the k th period is determined by taking the maximum from the precision bounds of all nodes in theInput:myid:my storage id number/*Get localδ.*/1measure u(k)from monitor;2calculateδmyid(k)from local controller;3foreach peer node d in R−˘d myid¯do4/*Exchange localδs.*/5/*Use piggyback to save communication cost.*/ 6sendδmyid(k)to d;7receiveδi(k)from d;8end9/*Get theﬁnal globalδ.*/10δglobal(k)=max(δi(k)),where i∈R;11。

上海市风华中学2024-2025学年高三上学期9月阶段测试英语试题

上海市风华中学2024学年度第一学期高三年级英语阶段测试（2024.9）（满分140分考试时间：120分钟）第Ⅰ卷I. Listening ComprehensionSection A 10%Directions: In Section A, you will hear ten short conversations between two speakers. At the end of each conversation, a question will be asked about what was said The conversations and the questions will be spoken only once. After you hear a conversation and the question about it, read the four possible answers on your paper, and decide which one is the best answer to the question you have heard.1. A. She has no appetite at all. B. She wants to dine out.C. She is too tired to go out.D. She prefers to cook at home.2. A. 6 pounds B. 7 pounds. C. 8 pounds. D. 9 pounds.3. A. At the professor's office. B. In the bookstore.C. In the library.D. In the laboratory.4. A. Because something went wrong with his car.B. Because his car was broken in an accident.C. Because he wanted to take a walk for a rest.D. Because he was stuck in a traffic jam.5. A. The morning flight. B. The afternoon flight.C. The evening flight.D. The midnight flight.6. A. She is not interested in going camping with him.B. She wants the man to stay at home with her.C. She thinks the man needs to have a good rest.D. She thinks the man should prepare for the exams.7. A. Some major revisions are needed. B. It should be revised by a tutor.C. Only a few changes should be made.D. The draft needs no revision at all.8 A. He is going away for a while. B. He worked hard to earn money.C. He did very well in the exam.D. He can't wait to have a rest.9. A. He forgot to bring his own camera. B. He is not good at taking pictures.C. He cannot take a photo with the camera.D. He doesn't know how to use the camera.10. A. She was interrupted by a visiting friend. B. She didn't come back until midnight.C. She stayed up late for the final exam.D. She visited her friend instead of studying.Section BDirections: In Section B, you will hear two short passages, and you will be asked several questions on each of the passages. The passages will be read twice, but the questions will be spoken only once. When you hear a question, read the four possible answers on your paper and decide which one would be the best answer to the question you have heard.Questions 11 through 13 are based on the following passage.11. A. They can maintain their body temperature stable.B. They conserve enough energy before the long sleep.C. They can keep their heart beat at a regular rate.D. They have their weight increased to the maximum.12. A. By staying in bidıng places and eating little.B. By seeking extra food and warm shelter.C. By growing thicker hair to stay warm.D. By storing enough food in advance.13. A To stay safe. B. To save energy. C. To get more food. D. To protect the young. Questions 14 through 16 are based on the following passage.14. A. Four to six hours. B. Six to nine hours.C. Around eight hours.D. More than eight hours.15. A. They may not be able to concentrate well.B. They may get the feeling of being drunk.C. They may suffer from high blood pressure.D. They may lose weight easily in a short period of time.16. A. Military people are used to being deprived of sleep.B. Training can make people sleep less and suffer less.C. People can bank sleep by sleeping more beforehand.D. Sleeping earlier than usual makes people sleep less.Questions 17 through 20 are based on the following passage.17. A. Double Eleven sales in 2021.B. Unreliable factors of online shopping.C. Key points of Taobao's success ı n sales.D. Advantages and disadvantages of online shopping.18. A. People who are good at doing business.B. People who work seven days a week.C. People who have very busy schedules.D. People who dislike telephone shopping.19. A. Consumers can save a lot of time.B. It provides round- the- clock service.C. People can buy things without leaving their homes or offices.D. The quality of the product is the same as what is described online.20. A. Inferior quality. B. Various retailers.C. Efficient sales return.D. Convenient delivery.II. Grammar and VocabularySection ADirections: After reading the passages below, fill in the blanks to make the passages coherent and grammatically correct. For the blanks with a given word, fill in each blank with the proper form of the given word; for the other blanks, use one word that best fits each blankAI Weather Forecasting Can't Replace Humans- YetAs Hurricane Lee was curving（呈曲线）northward to the west of Bermuda in mid- September of last year, forecasters were busily consulting weather models and data from hurricane- hunter aircraft to calculate （21）________ the dangerous storm was likely to make landfall （着陆）: New England or farther east, in Canada. The sooner the meteorologists（气象学家）could do so, the earlier they could warn those in the path of damaging wind gusts and fierce storm surges By six days ahead of landfall, it was clear that Lee （22）________（follow）the eastward path, and warnings were issued, accordingly. But （mother tool- an experimental AI model called GraphCast-（23）________（mate）that outcome accurately three whole days before the forecasters' traditional models.GraphCast's prediction is a window into AI's potential （24）________（improve）weather forecasts. But whether it is a forecaster of a true sea change in the field or will simply become one of many tools （25）________ human forecasters consult to determine which way the winds will blow is still up in the air.GraphCast, developed by Google DeepMind, is the latest of several AI weather models （26）________（release）in recent years. Google's Metnet, first introduced in2020, is already being used in products such as the company's “now cast” in its weather app. All are advertised as having an accuracy that is comparable with or higher than（27）________ on the best non-AI forecasting computer models and have caused a sensation in meteorology, with GraphCast （28）________（cause）the most significant stir so far.The DeepMind research team had put GraphCast through its paces by feeding it historical weather data to see if it could accurately “predict” what happened. The study showed the AI performed equal to or even better than the gold standard.Yet （29）________ GraphCast becomes probabilistic-- and even if the model's resolution improves and the AI becomes more accurate in its forecasts of rain and storm intensity - modeling remains just a single component of the weather- prediction pipeline, says Hendrik Tolman, senior adviser for advanced modeling systems at the NWS. However, every expert described GraphCast and other Al models as additional devices in their tool kit. If AI （30）________produce accurate forecasts quickly and cheaply, there's no reason not to begin using it together with existing methods.But will there be a world where AI models replace physics- based models— and people -- in the future? Forecasts suggest there's little chance.Section BDirections: Complete the following passage by using the words or phrases in the box. Each word or phrase can only be used once. Note that there is one word or phrase more than you need.Alzheimer's Drug Approved Despite Fierce DebateThe -U. S. Food and Drug Administration （FDA）recently approved the drug Aduhelm, produced by Americanbiotechnology company Biogen with Japan's Eisai Co., to treat patients with Alzheimer's（老年痴呆症）disease. The approval was based on study results showing that the drug seemed“（31）________ likely” to benefit Alzheimer's patients, the FDA said.The decision, which could （32）________ millions of Alzheimer's patients and their families, has sparked disagreements among medical researchers. While the drug was shown to be effective in slowing the mental decline in patients' suffering from the disease, it was not proven to be effective in （33）________ its effects, the Associated Press reported. citing a study. The rate of mental decline in patients that had been administered Aduhelm was slowedby 22 percent when compared to patients who had received a placebo（安慰剂）. But even given these results, on a test that is conducted to evaluate the cognitive and （34）________ abilities of a patient, patients who were administered Adubelm only showed an increase of 0.39 in their- scores. And it's unclear how such metrics （度量标准）translate into practical benefits, like greater（35）________ or the ability to recall important details.The FDA's review of the drug has become a flashpoint in （36）________ debates over standards used to evaluate therapies for hard- to- treat conditions. On one side, groups representing Alzheimer's patients and their families sayany new therapy - even one of small benefit -deserves approval. But many experts ward that（37）________ the drug could set a dangerous example by opening the door to treatments of questionable benefit.Alzheimer's is an irreversible, （38）________ brain disorder that slowly attacks areas of the brain that are essential to memory, reasoning, communication, and basic daily tasks. In the final stages of the disease, the patientswill lose the ability to（39）________Science doesn't fully understand what causes Alzheimer's, but there's broad agreement that the brain plaque（斑点）that is being （40）________ by Aduhelm is one of the contributing factors. Evidence suggests family history, education, and chronic conditions like heart disease may all play a role. “This is a sign of hope but not the final answer,” said Dr. Richard Hodes, director of the U. S. National Institute on Aging.III. Reading ComprehensionSection ADirections: For each blank in the following passage there are four words or phrases marked A, B, C and D. Fill in each blank with the word or phrase that best fits the context.Some people like to read the instructions from start to finish before they take action while others study the diagrams and then jump right in. This 41 for one approach over another when learning new information is not uncommon. Indeed, the notion that people learn in different ways is such a universal belief in American culture that there is a thriving industry dedicated to 42 learning styles and training teachers to meet the needs of different learners.Just because a notion is popular, 43 , doesn't make it true. A recent review of learning styles found evidence to clearly support the idea that outcomes are 44 when instructional techniques align with （匹配）individuals' learning styles. Most previous investigations on learning styles focused on classroom learning, and assessed whether instructional style 45 outcomes for different types of learners. But is the 46 really where most of the serious learning occurs? Some might argue that, in this era of flipped classrooms and online course materials, students 47 more of the information on their own. That might explain why instructional style in the classroom matters little. It also 48 the possibility that learning styles do matter. Perhaps a 49 betweenstudents' individual learning styles and their study strategies is the key to ideal outcomes.To explore this 50 , researchers asked students enrolled in an anatomy class （解剖课）to complete an online learning styles assessment, answer questions about their study strategies and report details about the 51 they used outside of class（e. g. flash cards, review of lecture notes, anatomy coloring books）.Scores suggested that most students used multiple learning styles, but that no particular style 52 better outcomes than another. The focus in this study, however, was not on whether a particular learning style was more53 . Despite knowing their own, self- reported learning preferences, nearly 70% of students 54 to employ study techniques that supported those preferences. Given the popular belief that learning styles matter, and the fact that many students 55 poor academic performance on the lack of a match between their learning style and teachers' instructional methods, one might expect students to rely on techniques that support their personal learning preferences when working on their own.41. A. preference B. tendency C. phenomenon D. practice42. A. identifying B. exposing C. revealing D. establishing43. A. therefore B. moreover C. however D. instead44. A. best B. acceptable C. disappointing D. undesirable45. A. impacted B challenged C. confirmed D. supported46. A. network B. classroom C. school D. lecture47. A. require B. collect C. master D. demand48. A. limits B eliminates C examines D. raises49. A. comparison B. link C. balance D. match50. A. issue B. possibility C. field D. proposal51. A. equipment B. techniques C. notebooks D. assistance52. A. originated in B. resulted from C. resulted in D. took over53. A. important B. advantageous C meaningful D popular54. A. failed B. managed C. struggled D. attempted55. A. count B. concentrate C. blame D. conductSection BDirections: Read the following four passages. Each passage is followed by several questions or unfinished statements. For each of them there are four choices marked A, B, C and D. Choose the one that fits best according to the information given in the passage you have just read.（A）I was sure that I was to be killed. I became terribly nervous. I fumbled （摸索）in my pockets to see if there were any cigarettes, which had escaped their search. I found one and because of my shaking hands, I could barely get it to my lips. But I had no matches; they had taken those. I looked through the bars at my jailer. He did not make eye contact with me. I called out to him, “Have you got a light?” He looked at me, shrugged and came over to light my cigarette. As he came close and lit the match, his eyes unconsciously locked with mine. At that moment, I smiled. I don't know why I did that. Perhaps it was nervousness, perhaps it was because, when you get very close, one to another, it is very hard not to smile. In any case, I smiled. In that instant, it was as though a spark jumped across thegap between our two hearts, our two human souls. I know he didn't want to, but my smile leaped through the bars and generated a smile on his lips, too. He lit my cigarette but stayed near, looking at me directly in the eyes and continuing to smile.I kept smiling at him, now aware of him as a person and not just a jailer. And his looking at me seemed to have a new dimension too.“Do you have kids?” he asked.“Yes, here, here.”I took out my wallet and nervously fumbled for the pictures of my family. He, too, took out the pictures of his family and began to talk about his plans and hopes for them. My eyes filled with tears. I said that I feared that I'd never see my family again, never have the chance to see them grow up. Tears came to his eyes, too. Suddenly, without another word, he unlocked my cell and silently led me out. Out of the jail, quietly and by back routes, out of the town. There, at the edge of town, he released me. And without another word, he turned back toward the town.My life was saved by a smile, yes, the smile- the unaffected, unplanned, natural connection between people. I really believe that if that part of you and that part of me could recognize each other, we wouldn't be enemies. We couldn't have hate or envy or fear.56. The underlined sentence indicates that the author and the jailer started to have a ________ conversation.A. less impersonalB. more intenseC. less formalD. more friendly57. Which is true based on the first paragraph?A. My hands were shaking because of fear.B. The jailer was going to shoot me.C. I smile8 because I had to beg for life.D. He smiled to me because he wanted to.58. Their eyes were filled with tears because they both ________.A. took out the pictures of their familiesB. missed their families far awayC. had plans and hopes for futureD. feared that they would die59. How does a smile succeed in saving the author's life?A. By asking for the jailer to light a cigarette.B. By planning for an exchange of family pictures.C. By establishing natural connection between people.D. By hiding the human feelings of hate, envy or fear.（B）CareersHome>> How To ApplyFAQs on preparing your ApplicationQ: Should I target my Application to a specific Job Opening（JO）?A: Yes. Naturally, a customized cover note will also help you focus on the key aspects of your Application that relateto the JO, but it is also in your interest to target the Application according to the responsibilities and competencies of the position.Q: What's the difference between duties and achievements?A: Duties describe the specific responsibilities of your job. They accurately reflect what you are doing or have done in each of your previous jobs. In other words, it is the “what you do” of your job. Achievements describe in specific terms “how well” you did in your job.Q: Many of my achievements are team- based, how do I draft them in my Application?A: You should include your team- based achievements in your Application. Indicate that you were part of a team, and describe your specific role in reaching the goal.FAQs on general Application guidelinesQ: Can I save my Application?A: Yes. You should save your Application when you make changes and/ or update it. It is recommended that you save different versions of your Application in Word format and then edit the Application online according to the post for which you are applying.Q: Can I update my Application to apply for a new JO?A: Yes. Each time you apply for a new JO, we recommend that you review your Application and update it , if appropriate, or target it to better reflect your suitability for the new JO. Your updates will not affect the content of Applications previously submitted against other JOs.Q: Must I use up all the available characters in each section of my Application?A: No. In fact, doing so may result in an unnecessary lengthy Application. Unless you have an enormous range of experiences, there is no reason to use up all the space given. Applicants are encouraged to list their duties and achievements in a clear and brief manner.60. Which of the following descriptions best shows your achievements?A. I've developed various interests, ranging from oil painting to designing model.B. I'm good at creating proposals for new product ideas aimed at a specific market.C. I'm in charge of the clearance, production and distribution of information material.D. I succeeded in directing a video presentation, assisting our group to win the first prize.61. If you want to apply for another JO, you'd better __________.A. target your focus on your interest in the JOB. save your latest application in Word formatC. Serape one application with all your competenciesD. update your application to match new requirements62. Applicants are expected to __________ in their applications.A. introduce what JOs they have previously applied toB. list the greatest achievements they have made in detailC. give key information about their experiences and achievementsD provide the results of their tests, assessments and examinations（C）Atlantis is the legendary island that sank beneath the waves in the distant past, taking down with it an advanced civilization. Is it possible that we will ever find it? Or, more importantly did it even exist?The short answer to both: No. All available evidence indicates that the philosopher Plato, sometime around 360 B.C., invented the island nation to illustrate a point about the dangers of aggressive imperialism（势力扩张）. In Plato's telling, Atlantis was no utopia. Rather, it was a centrist to an idealized version of Athens from long before Plato's time. This ancient Athens was very similar to Plato's notion of the ideal state. Plato laid out the details for what such a state would look like in his famous work, I he Republic. It should be small and virtuous. The residents of Atlantis, on the other hand, were eventually “filled with an unjust lust for possessions and power," according to Plato's character who described the island.In Plato's texts, Atlantis was “larger than Libya and Asia combined,”（which, in Plato's time. would have referred to modern- day northern Africa and over half of Turkey）. It was situated in the Atlantic Ocean, somewhere outward from the Strait of Gibraltar. It's a landmass large enough that, if it really existed somewhere underwater in the Atlantic, it would certainly appear on sonar maps of the ocean floor.So how did Atlantis come to represent a lost utopic civilization? For that, you can mostly blame （or thank）Ignatius Donnelly. In 1882, the former U. S. Congressman published Atlantis. The Antediluvian World. The book laid out 13 hypotheses, centered on the idea that Atlantis had truly existed, and indeed represented a place “where early mankind dwelt for ages in peace and happiness. According to Donnelly, Atlantis was the original source of many ancient civilizations around the world. If one followed the clues in Plato's writing, Donnelly believed, Atlantis could be found. He was inspired by a remarkable discovery in the early 1870s. An amateur archaeologist claimed to have unearthed the legendary city of Troy based on Homer's The Iliad. If Troy, long thought to be fictional, was real, why shouldn't Atlantis be, too?Donnelly was certain of his theory, predicting that hard evidence of the sunken city would soon be found, and that museums around the world would one day be filled with artifacts from Atlantis. Yet about 140 years have passed without a trace of evidence. The Atlantis legend has been kept alive, fueled by the public's imagination and fascination with the idea of a hidden. long- lost utopia. Yet the “lost city of Atlantis was never lost; it is where it always was: in Plato's books.63. What can we learn about Plato?A. He predicted that Atlantis would be' destroyed by aggressive imperialism.B. He was inspired by utopia to gradually form the notion of the ideal state.C. He created the setting in which residents of Atlantis were not virtuous.D. He witnessed Atlanteans' pursuit of an unjust lust for possessions and power.64. Homer's The Iliad is mentioned ___________.A. to demonstrate the actual existence of the legendary city of Troy.B. as indirect evidence of the credibility of Plato's account of Atlantis.C. because it is a great piece of fictional writing about an ancient legend.D. because it contains many clues about the legendary city of Troy.65. According to the passage, Atlantis was ___________.A. a long- lost small utopia with many virtuous residents.B. a large landmass situated in the Atlantic Ocean.C. the original source of many ancient civilizations.D. Plato's invention against which to highlight his ideal.66. Which of the following is the best title for this passage?A. Plato, Atlantis and How the City Collapsed and Finally Got LostB Plato Told a Lie, and Ignatius Donnelly was to Blame for ItC. The History, Legends, and Evidence of the Lost City of AtlantisD. Where Is the Lost City of Atlantis — and Does It Even Exist?Section CDirections: Read the passage carefully. Fill in each blank with a proper sentence given in the box. Each sentence can be used only once. Note that there are two more sentences than you need.Bringing Light to the Darkness with Crisco ArtMost paintings are best enjoyed in galleries with good lighting. But an Italian artist who goes by the name Crisco is changing the way we look at paintings with a new approach: glow （发光）- in- the- dark paint.Crisco's paintings are beautiful in normal lighting, but it is when the lights go down that they really come alive. （67）_________ His art mostly shows landscapes. Trees, horizons, and especially starry skies come alive with the glow of his paints. At the center of most of his work, there is often a human or animal figure. The figure may be just a shadow surrounded by the glowing colors, but it often appears to be the source of the light. （68）_________ Instead, they are all bright pictures of hope, life, wonder, and growth. They are Crisco's way o t adding a little light to the world.Crisco's full name is Cristoforo Scorpiniti.（69）_________ Instead of letting a negative experience get the best of him, he threw himself into a new pursuit: art. According to Crisco, he paints with glowing colors to inspire hope. Though his paintings often show night scenes that look good in the dark, Crisco does not focus on the darkness. Instead, he uses his paintings to express positivity by creating light in the darkness.A lot of his best work has come out of just painting what he felt at the time without any plan or structure.（70）_________ With over half a million followers on Instagram, Crisco is already popular on social media for his unique paintings. He'll surely only get more famous in the future for his inspiring paintings that beautifully mix darkness and light.IV. Summary Writing71. Directions: Read the following passage. Summarize the main idea and the main point（s）of the passage in no more than 60 words. Use your own words as far as possible.Are EV Really Environmentally Friendly?Many consumers are opting for an electric vehicle （EV）or plug- in hybrid electric vehicle （PHEV）to replace their polluting gas- powered cars. These electrified vehicles are rising to popularity on the premise of environmental conservation and eliminating the need for harmful emissions. There are a couple of things. however, to consider before concluding that EVs are the most environmentally friendly option for consumers.Where do electric cars get their energy? Although EVs create no emissions on board, they typically draw power from lithium- ion batteries. These batteries require charging, either at home or via a publicly accessible charging station. Since EV charging infrastructure is mainly reliant on the power grid - specifically, the grid draws power from plants like coal plants - although your EV does not produce any harmful emissions as you drive it, burning fossil fuels is involved in fueling it. Moreover, temperature extremes like excessive coldness or heat can dramatically reduce lithium- ion battery life. Carnegie Mellon University's Department of Engineering and Technology says that the most extreme cases of coldness will compromise efficiency by as much as 40%. The decreased efficiency is an issue if the power stored in the battery packs of EVs is sourced from fossil fuel- burning.Besides the power source, metal s such as lithium and cobalt are wrapped up in environmentally and socially questionable processes, too. One of the first environmental issues lithium batteries pose is how to dispose of them properly. In an average battery recycling plant, all parts of the battery are shredded into a powder using a mechanical shredder and then either melted or dissolved into acid —recycling lithium batteries isn't as simple. Lithium batteries are typically made up of a mix of different elements including cobalt, nickel, manganese and iron —cobalt especially known to be a hazardous substance. In addition, most studies associate lithium mining in South America from salt brine with salinization（盐化）of freshwater that the locals need to survive. Since the mineral contains dangerous substances, the mining process also contaminates the local water basins. So, lithium extraction exposes the local ecosystems to poisoning and other related health problems.V. TranslationDirections: Translate the following sentences into English, using the words given in the brackets.72.他那种急于求成的心态让他无缘冠军宝座。

基础模型与顺序决策的问题、方法、及应用

基础模型与顺序决策的问题、方法、及应用1.基础模型是指具有固定结构和参数的数据模型。

The basic model refers to a data model with a fixed structure and parameters.2.顺序决策是一种根据先后顺序进行决策的方法。

Sequential decision-making is a method of making decisions based on the order of events.3.基础模型可以用于对数据进行简单的描述和分析。

Basic models can be used to simply describe and analyze data.4.顺序决策需要考虑未来的可能性和不确定性。

Sequential decision-making requires consideration of future possibilities and uncertainties.5.基础模型通常假设数据之间的关系是确定的。

Basic models typically assume that the relationships between data are deterministic.6.顺序决策需要考虑不同决策可能带来的不同结果。

Sequential decision-making requires consideration of the different outcomes that different decisions might bring.7.基础模型的应用范围广泛，可以用于各种领域的数据分析。

Basic models have a wide range of applications and can be used for data analysis in various fields.8.顺序决策可以在金融、管理和工程等领域中得到应用。

感兴趣的可靠性书籍

感兴趣的可靠性书籍已有1590次阅读2015-9-2210:27|个人分类:可靠性技术|系统分类:科研笔记|关键词:可靠性分析环境试验设备可靠性书籍一直从事可靠性方面的工作，看过几十本关于环境试验中文版本的标准，也参与起草过2个国标的编写。

近2年稍微时间比较充裕，打算把以下书籍浏览一遍，任务可不轻，有很多书可能都买不到或者借不到。

如果大家有好的可靠性图书也欢迎推荐给我。

1.Reliability Engineering Handbook(Volume1)–Dimitri Kececioglu2.Reliability Engineering Handbook(Volume2)–Dimitri Kececioglu3.Reliability&Life Testing Handbook,Volume1–Dimitri Kececioglu4.Reliability&Life Testing Handbook,Volume2–Dimitri Kececioglu5.Robust Engineering Design-By-Reliability with EMphasis on MEchanical Components and Structural Reliability,Vol.1–Dimitri Kececioglu6.Environmental Stress Screening:Its Quantification,Optimization and Management–Dimitri Kececioglu7.The New Weibull Handbook Fifth Edition,Reliability and Statistical Analysis for Predicting Life,Safety,Supportability,Risk,Cost and Warranty Claims8.Maintenance and Reliability Best Practices9.Software Reliability:Measurement,Prediction,Application10.Software Reliability Engineering:More Reliable Software Faster and Cheaper2nd Edition11.Automotive Electronics Reliability(Progress in Technology)12.Applied Reliability–Third edition13.Achieving System Reliability Growth Through Robust Design and Test14.电子元器件应用手册（参考书）15.轨道列车可靠性、可用性、维修性和安全性16.动车组结构可靠性与动力学17.可靠性工程与管理实践–怎样提高产品可靠性18.疲劳强度设计19.统计学–科学与工程应用20.概率统计21.可靠性设计大全22.风力机可靠性工程23.耐热钢持久性能的统计分析及可靠性预测24.故障诊断、预测与系统健康管理（培训课时看过）25.现代机械工程设计–全寿命周期性能与可靠性26.系统可靠性设计与分析27.可靠性与维修性工程概论28.可靠性工程数学29.结构可靠性理论与应用30.电子元器件可靠性设计31.产品可靠性、维修性及保障性手册32.数控机床性能分析及可靠性设计技术33.液压系统可靠性工程34.可靠性数据分析教程（看过）35.漫画玩转统计学36.软件可靠性工程37.高可靠性航空产品试验技术38.系统可靠性评定方法研究39.空间运载器的可靠性保证40.机械系统设计初期的可靠性模糊预计与分配41.MEMS可靠性42.高加速寿命试验与高加速应力筛选（此书翻译质量较差，建议大家不要购买）43.Accelerated Reliability Engineering—HALT and HASS44.Contributions to Hardware and Software Reliability45.Sensor Performance and Reliability46.Reliability Toolkit:Commercial Practices Edition–A Practical Guide for Commercial Products and Military Systems Under Acquisition Reform（已阅读）47.Engineering Design Reliability Handbook48.Reliability Improvement with Design of Experiment,Second Edition49.Design for Reliability(Quality and Reliability Engineering Series)50.Reliability Data Analysis With Excel and Minitab51.Effective FMEAs:Achieving Safe,Reliable,and Economical Products and Processes using Failure Mode and Effects Analysis52.Global Vehicle Reliability:Prediction and Optimization Techniques（已阅读）53.电子封装技术丛书：电子封装技术与可靠性54.大功率电站汽轮机寿命预测与可靠性设计55.汽车可靠性工程基础56.航天器机构及其可靠性57.腐蚀试验方法及监测技术（已阅读）58.机械可靠性：理论·方法·应用59.大容量电站锅炉可靠性与寿命的设计及评定60.普通高等教育十五国家级规划教材：汽车可靠性技术61.Accelerated Reliability and Durability Testing Technology62.Design and Analysis of Accelerated Tests for Mission Critical Reliabilitypressors:How to Achieve High Reliability&Availability64.Product Warranty Handbook65.Electronic Derating for Optimum Performance66.Automotive Electronics Reliability,Volume267.Vibration Spectrum Analysis68.可靠性工程（第2版）69.现代机械设计手册·单行本：疲劳强度与可靠性设计70.电子组装工艺可靠性71.机械可靠性工程（已阅读）72.Warranty Cost Analysis73.Reliability of Electronic Components:A Practical Guide to Electronic Systems Manufacturing74.Long-Term Non-Operating Reliability of Electronic Productsponent Reliability for Electronic Systems76.THE RELIABILITY HANDBOOK VOLUME1(NATIONAL SEMICONDUCTOR CORPORATION)77.At&t Reliability Manual78.Reliability of Large Systems79.Understanding Measurement:Reliability(Understanding Statistics)80.15Most Common Obstacles to World-Class Reliability:A Roadmap for Managers81.Reliability Theory With Applications to Preventive Maintenance82.Early Prediction models for software reliability83.Reliability Assurance for Medical Devices,Equipment and Software84.Reliability Assessment:A Guide to Aligning Expectations,Practices,and Performance85.Introduction to the Physics of Materials86.振动信号的现代分析技术与应用87.振动冲击及噪声测试技术(第二版)88.Vibration Spectrum Analysis89.Random Vibration in Perspective90.Reliability-Based Design91.冲击与振动手册（第5版）92.Ensuring Software Reliability93.Handbook of Reliability Engineering and Management2/E94.Reliability:Modeling,Prediction,and Optimization95.云计算实战:可靠性与可用性设计96.可靠性物理与工程:失效时间模型97.功率半导体器件:原理、特性和可靠性98.A Minimal-Mathematics Introduction to the Fundamentals of Random Vibration& Shock Testing:Measurement,Analysis and Calibration as Applied to Halt99.Reliability and Degradation of Semiconductor Lasers and LEDs100.Reliability and Fault Tree Analysis:Theoretical and Applied Aspects of System Reliability and Safety Assessment101.Fault Tree Analysis Primer Chinese Edition102.Reliability:For Technology,Engineering,and Management103.Methods for Statistical Analysis of Reliability and Life Data104.Reliability Engineering for Electronic Design105.Reliability physics(Volume6)106.Reliability Modelling:A Statistical Approach107.Introduction to Machinery Reliability Assessment108.Reliability and Validity in Qualitative Research109.Resistor Theory and Technology110.The Inductor Handbook:A Comprehensive Guide For Correct Component Selection In All Circuit Applications.Know What To Use When And Where.111.The Capacitor Handbook:A Comprehensive Guide For Correct Component Selection In All Circuit Applications.Know What To Use When And Where.112.THe Diode Handbook113.The Transistor Handbook114.The Resistor Handbook115.电容器手册116.Electronic Packaging:Design,Materials,Process,and Reliability117.Probability,Statistics,and Reliability for Engineers and Scientists,Second Edition 118.Reliability:For Technology,Engineering,and Management119.Reliability Engineering and Risk Assessment120.Reliability of RoHS-Compliant2D and3D IC Interconnects121.Reliability-Based Design in Civil Engineering122.Integrated Circuit Quality and Reliability,Second Edition123.Hydrosystems Engineering Reliability Assessment and Risk Analysis124.Digital Switching Systems:System Reliability and Analysis125.Reliability in Procurement and Use:From Specification to Replacement。

深圳读书月这10本经典英文绘本小学生必读

• Developing reading habits: By continuously reading picture books, primary school students can develop reading habits and independent reading abilities.
Books • Language Learning in English Picture
Books
2
目录
• Cross cultural awareness cultivation and international perspective expansion
• How parents guide their children to read English picture books
After reading, you can engage in related extended activities with your child, such as role-playing, drawing, etc.
Help children understand the difficulties and doubts in picture books through discussion.
2024/1/25
21
04
Language Learning in English Picture Books
Chapter
2024/1/25
22
Vocabulary accumulation and application
By reading classic English picture books, elementary school students can be exposed to rich vocabulary and commonly used expressions in daily life, such as animals, colors, numbers, family, school, and other related topics.

Protein Function Predict

汇报人
刘言
Sample of homologous sequences
MSA and Family features identification
Secondary structure prediction
Flod recognition and Secondary structure alignment
a crystal structure of another member of the same family to be available. （more than 25% of identical amino acids）
Comparative modeling of the investigated sequence
Check of physical-chemical contains
rotamers libraries, energy minimization, molecular dynamics,pdb loop databases
Binding pocket prediction within the best comparative model
mutagenesis and literature data about binding
Molecular based docking and virtual screening of chemical libraries in the proposed binding region
在蛋白质表面寻找那些与生物化学和细胞学功能有关的位点始终是一个问题。 • Sequence-based methods allow the identiﬁcation of a ligandbinding interaction motif. • Structure-based methods are reliant on homology and require

湍流长度尺度英文

湍流长度尺度英文Turbulence Length ScalesTurbulence is a complex and fascinating phenomenon that has been the subject of extensive research and study in the field of fluid mechanics. One of the key aspects of turbulence is the concept of turbulence length scales, which refers to the range of different-sized eddies or vortices that are present in a turbulent flow. These length scales play a crucial role in understanding and predicting the behavior of turbulent flows, and they have important implications in a wide range of engineering and scientific applications.The smallest length scale in a turbulent flow is known as the Kolmogorov length scale, named after the Russian mathematician and physicist Andrey Kolmogorov. This length scale represents the size of the smallest eddies or vortices in the flow, and it is determined by the rate of energy dissipation and the kinematic viscosity of the fluid. The Kolmogorov length scale is typically denoted by the Greek letter η (eta) and can be expressed as η = (ν^3/ε)^(1/4), where ν is the kinematic viscosity of the fluid and ε is the rate of energy dissipation.The Kolmogorov length scale is important because it represents the scale at which viscous forces become dominant and energy is dissipated into heat. Below this length scale, the flow is considered to be in the dissipation range, where the eddies are too small to sustain their own motion and are rapidly broken down by viscous forces. The Kolmogorov length scale is therefore a critical parameter in the study of turbulence, as it helps to define the range of scales over which energy is transferred and dissipated within the flow.Another important length scale in turbulence is the integral length scale, which represents the size of the largest eddies or vortices in the flow. The integral length scale is typically denoted by the symbol L and is a measure of the size of the energy-containing eddies, which are responsible for the bulk of the turbulent kinetic energy in the flow. The integral length scale is often determined by the geometry of the flow domain or the boundary conditions, and it can be used to estimate the overall scale of the turbulent motion.Between the Kolmogorov length scale and the integral length scale, there is a range of intermediate length scales known as the inertial subrange. This range is characterized by the presence of eddies that are large enough to be unaffected by viscous forces, but small enough to be unaffected by the large-scale features of the flow. In this inertial subrange, the energy is transferred from the large eddies to the smaller eddies through a process known as the energycascade, where energy is transferred from larger scales to smaller scales without significant dissipation.The energy cascade is a fundamental concept in turbulence theory and is described by Kolmogorov's famous 1941 theory, which predicts that the energy spectrum in the inertial subrange should follow a power law with a slope of -5/3. This power law relationship has been extensively verified through experimental and numerical studies, and it has important implications for the modeling and prediction of turbulent flows.In addition to the Kolmogorov and integral length scales, there are other important length scales in turbulence that are relevant to specific applications or flow regimes. For example, in wall-bounded flows, the viscous length scale and the boundary layer thickness are important parameters that can influence the turbulent structure and behavior. In compressible flows, the Taylor microscale and the Corrsin scale are also relevant length scales that can provide insight into the characteristics of the turbulence.The understanding of turbulence length scales is crucial for a wide range of engineering and scientific applications, including fluid dynamics, aerodynamics, meteorology, oceanography, and astrophysics. By understanding the different length scales and their relationships, researchers and engineers can better predict andmodel the behavior of turbulent flows, leading to improved designs, more accurate simulations, and a deeper understanding of the fundamental principles of fluid mechanics.In conclusion, turbulence length scales are a fundamental concept in the study of turbulent flows, and they play a crucial role in our understanding and modeling of this complex and fascinating phenomenon. From the Kolmogorov length scale to the integral length scale and the inertial subrange, these length scales provide valuable insights into the structure and dynamics of turbulence, and they continue to be an active area of research and exploration in the field of fluid mechanics.。

MCD12Q1_的user guide用户手册或manual说明

User G uide f or t he M ODIS L and C over T ype P roduct (MCD12Q1)Last U pdated: A ug 8, 20121. IntroductionLand c over p lays a m ajor r ole i n t he c limate a nd b iogeochemistry o f t he E arth system. A n i mportant u se o f g lobal l and c over d ata i s t he i nference o f p arameters that i nfluence b iogeochemical a nd e nergy e xchanges b etween t he a tmosphere a nd the l and s urface f or u se i n m odels a nd o ther g lobal c hange s cience a pplications. Examples o f s uch p arameters i nclude l eaf a rea i ndex, r oughness l ength, s urface resistance t o e vapotranspiration, c anopy g reenness f raction, v egetation d ensity, root d istribution, a nd t he f raction o f p hotosynthetically-‐active r adiation a bsorbed. The M ODIS L and C over T ype P roduct p rovides a s uite o f l and c over t ypes t hat support g lobal c hange s cience b y m apping g lobal l and c over u sing s pectral a nd temporal i nformation d erived f rom M ODIS. T he o bjective o f t his d ocument i s t o provide i nformation r elated t o t he C ollection 5 M ODIS L and C over T ype P roduct (MCD12Q1). I t i s n ot d esigned t o b e a s cientific d ocument. R ather, i t p rovides t hree main t ypes o f i nformation:1.An o verview o f t he M CD12Q1 a lgorithm a nd p roduct, a long w ith r eferencesto p ublished l iterature w here m ore d etails c an b e f ound.2.Guidance a nd i nformation r elated t o d ata a ccess a nd d ata f ormats, t o h elpusers a ccess a nd u se t he d ata.3.Contact i nformation f or u sers w ith q uestions t hat c annot b e a ddressedthrough i nformation o r w ebsites p rovided i n t his d ocument.2. O verview o f t he M CD12Q1 L and C over T ype P roductThe M ODIS L and C over T ype P roduct i s p roduced u sing a s upervised c lassification algorithm t hat i s e stimated u sing a d atabase o f h igh q uality l and c over t raining sites. T he t raining s ite d atabase w as d eveloped u sing h igh-‐resolution i magery i n conjunction w ith a ncillary d ata (Muchoney e t a l., 1999). T he s ite d atabase i s a “living” d atabase t hat r equires o n-‐going a ugmentation a nd m aintenance t o i mprove the t raining d ata a nd d etect m islabeled s ites o r s ites t hat h ave c hanged o ver t ime. MODIS d ata u sed i n t he c lassification i nclude a f ull y ear o f c omposited 8-‐day M ODIS observations. S pecific i nputs i nclude N ormalized B RDF-‐Adjusted R eflectance (NBAR; S chaaf e t a l., 2002) a nd M ODIS L and S urface T emperature (LST; W an e t a l., 2002) d ata. T hese f eatures a re p rovided t o t he c lassifier a s m onthly c omposites a nd annual m etrics (see F riedl e t a l., 2002; 2010).The c lassification i s p roduced u sing a d ecision t ree c lassification a lgorithm (C4.5; Quinlan 1993) i n c onjunction w ith a t echnique f or i mproving c lassification accuracies k nown a s b oosting (Freund 1995). B oosting i mproves c lassification accuracies b y i teratively e stimating a d ecision t ree w hile s ystematically v arying t he training s ample. A t e ach i teration t he t raining s ample i s m odified t o f ocus t he classification a lgorithm o n t he m ost d ifficult e xamples. T he b oosted c lassifier'sprediction i s t hen b ased u pon a n a ccuracy-‐weighted v ote a cross t he e stimatedclassifiers. T he i mplementation u sed h ere i s A daboost.M1 (Freund a nd S chapire, 1997), w hich i s t he s implest m ulti-‐class b oosting m ethod. B oosting h as b een s hownto b e a f orm o f a dditive l ogistic r egression (Friedman e t a l. 2000). A s a r esult, probabilities o f c lass m embership c an b e o btained f rom b oosting. T hese probabilities p rovide a m eans o f a ssessing t he c onfidence o f t he c lassification results a s w ell a s a m eans o f i ncorporating a ncillary i nformation i n t he f orm o f p rior probabilities t o i mproved d iscrimination o f c over t ypes t hat a re d ifficult t o s eparate in t he s pectral f eature s pace.Using t his a pproach, t he M ODIS L and C over T ype a lgorithm i ngests M ODIS t raining data f or a ll s ites i n t he t raining d atabase, e stimates b oosted d ecision t rees b ased o n those d ata, a nd t hen c lassifies t he l and c over a t e ach M ODIS l and p ixel. F ollowing the c lassification a s et o f p ost-‐processing s teps i ncorporate p rior p robability knowledge a nd a djust s pecific c lasses b ased o n a ncillary i nformation. F or m ore specific i nformation a nd c omplete d etails r elated t o t he M ODIS L and C over T ype algorithm, t he r eader i s r eferred t o t he f ollowing k ey r eferences:•Friedl e t a l. (1997)•Friedl e t a l. (1999)•McIver a nd F riedl (2001)•McIver a nd F riedl (2002)•Friedl e t a l. (2002)•Friedl e t a l. (2010)Full c itations t o e ach o f t hese p apers a re p rovided b elow.3. P roduct O verview a nd S cience D ata S etsThe M ODIS L and C over T ype P roduct s upplies g lobal m aps o f l and c over a t a nnualtime s teps a nd 500-‐m s patial r esolution f or 2001-‐present. T he p rimary l and c overscheme i s p rovided b y a n I GBP l and c over c lassification (Belward e t a l., 1999; Scepan, 1999; F riedl e t a l., 2002; F riedl e t a l., 2010). F or e ase o f u se b y t he community, a n umber o f o ther c lassification s chemes a re a lso p rovided, i ncluding the U niversity o f M aryland c lassification s cheme (Hansen e t a l., 2000), t he B iome classification s cheme d escribed b y R unning e t a l. (1994), t he L AI/fPAR B iome scheme d escribed b y M yneni e t a l. (1997), a nd t he p lant f unctional t ype s cheme described b y B onan e t a l. (2002). I n a ddition, a n a ssessment o f t he r elative classification q uality (scaled f rom 0-‐100) i s p rovided a t e ach p ixel, a long w ith quality a ssurance i nformation a nd a n e mbedded l and/water m ask.The m ost r ecent v ersion o f t he M ODIS L and C over T ype P roduct i s C ollection 5.1,which i ncludes a djustments f or s ignificant e rrors t hat w ere d etected i n C ollection 5 of t he M CD12Q1 p roduct. T his v ersion i s a vailable o n t he L and P rocesses D AAC a nd is t he r ecommended v ersion f or u sers. E ssential i nformation r equired f or a ccessing and u sing t hese d ata i nclude t he f ollowing:•Overview o f d ata s et c haracteristics (temporal c overage, s patial r esolution, image s ize, d ata t ypes, e tc.).•Science d ata s ets i ncluded i n t he M ODIS L and C over T ype P roduct, a nd t heir associated d efinitions.•Information a nd s pecifications r elated t o t he M ODIS L and C over T ype Q A Science d ata s et.Up-‐to-‐date i nformation r elated t o e ach o f t hese t opics i ncluding s cience d ata s ets, data f ormats, a nd q uality i nformation a re a vailable f rom t he L and P rocesses D AAC at t he f ollowing U RL:https:///products/modis_products_table/mcd12q13.1. D ata F ormats a nd P rojectionMODIS d ata a re p rovided a s t iles t hat a re a pproximately 10° x 10° a t t he E quator using a s inusoidal g rid i n H DF4 f ile f ormat. I nformation r elated t o t he M ODIS sinusoidal p rojection a nd t he H DF4 f ile f ormat c an b e f ound a t:•MODIS t ile g rid: h ttp:///MODLAND_grid.html•MODIS H DF4: h ttp:///products/hdf4/3.2. A ccessing a nd A cquiring D ataMCD12Q1 d ata c an b e a cquired f rom t he L and P rocesses D istributed A ctive A rchive Center (https:///get_data). T here a re m ultiple p ortals f or downloading t he d ata. R everb i s t he e asiest t o u se a nd d oes n ot r equire a u ser account, b ut y ou o nly h ave t he o ption t o d ownload t he d ata i n i ts o riginal p rojection and H DF f ormat. T he M RTWeb p ortal e nables m ore a dvanced o ptions s uch a s reprojection, s ubsetting, a nd r eformatting b ut d oes r equire a u ser a ccount.4. C ontact I nformationProduct P I: M ark F riedl (friedl@)Associate t eam m ember a nd c ontact f or u sers: D amien S ulla-‐Menashe(dsm@)5. R eferences C ited1.Belward, A. S., E stes, J. E., & K line, K. D. (1999). T he I GBP-‐DIS G lobal 1-‐km L and-‐Cover D ata S et D ISCover: A P roject O verview. P hotogrammetric E ngineering a nd Remote S ensing, 65, 1013-‐1020.2.Bonan, G. B., O leson, K. W., V ertenstein, M., L evis, S., Z eng, X. B., & D ai, Y. (2002).The l and s urface c limatology o f t he c ommunity l and m odel c oupled t o t he N CAR community l and m odel. J ournal o f C limate, 15, 3123-‐3149.3.Freund, Y. (1995). B oosting a w eak l earning a lgorithm b y m ajority. I nformationand C omputation, 121(2), 256-‐285.4.Freund, Y., & S chapire, R. E. (1997). A d ecision-‐theoretic g eneralization o f o n-‐linelearning a nd a n a pplication t o b oosting. J ournal o f C omputer a nd S ystem S ciences, 5(1), 119-‐139.5.Friedl, M.A., & B rodley, C.E. (1997). D ecision t ree c lassification o f l and c overfrom r emotely s ensed d ata. R emote S ensing o f E nvironment, 61, 399-‐409.6.Friedl, M.A., B rodley, C.E., & S trahler, A.H. (1999). M aximizing l and c overclassification a ccuracies a t c ontinental t o g lobal s cales. I EEE T ransactions o nGeoscience a nd R emote S ensing, 37, 969-‐977.7.Friedl, M. A., M cIver, D. K., H odges, J. C. F., Z hang, X. Y., M uchoney, D., S trahler, A.H., W oodcock, C. E., G opal, S., S chneider, A., C ooper, A., B accini, A., G ao, F., &Schaaf, C. (2002). G lobal l and c over m apping f rom M ODIS: a lgorithms a nd e arly results. R emote S ensing o f E nvironment, 83, 287-‐302.8.Friedl, M. A., S ulla-‐Menashe, D., T an, B., S chneider, A., R amankutty, N., S ibley, A.,& H uang, X. (2010). M ODIS C ollection 5 g lobal l and c over: A lgorithm r efinements and c haracterization o f n ew d atasets. R emote S ensing o f E nvironment, 114, 168-‐182.9.Friedman, J., H astie, T., & T ibshirani, R. (2000). A dditive l ogistic r egression: Astatistical v iew o f b oosting. T he A nnals o f S tatistics, 28(2), 337-‐374.10.Hansen, M. C., D eFries, R. S., T ownshend, J. R. G., & S ohlberg, R. (2000). G loballand c over c lassification a t t he 1km s patial r esolution u sing a c lassification t ree approach. I nternational J ournal o f R emote S ensing, 21, 1331-‐1364.11.Muchoney, D., S trahler, A., H odges, J., & L oCastro, J. (1999). T he I GBP D ISCoverConfidence S ites a nd t he S ystem f or T errestrial E cosystem P arameterization: Tools f or V alidating G lobal L and C over D ata. P hotogrammetric E ngineering a nd Remote S ensing, 65(9), 1061-‐1067.12.McIver, D. K., & F riedl, M. A. (2001). E stimating p ixel-‐scale l and c overclassification c onfidence u sing n on-‐parametric m achine l earning m ethods. I EEE Transactions o n G eoscience a nd R emote S ensing, 39(9), 1959-‐1968.13.Mciver, D. K., & F riedl, M. A. (2002). U sing p rior p robabilities i n d ecision-‐treeclassification o f r emotely s ensed d ata. R emote S ensing o f E nvironment, 81, 253-‐261.14.Myneni, R. B., N emani, R. R., & R unning, S. W. (1997). E stimation o f g lobal l eafarea i ndex a nd a bsorbed P AR u sing r adiative t ransfer m odel. I EEE T ransactions on G eoscience a nd R emote S ensing, 35, 1380-‐1393.15.Quinlan, J. R. (1993). C4.5: P rograms f or M achine L earning. S an M ateo, C A:Morgan K aufmann.16.Running, S. W., L oveland, T. R., & P ierce, L. L. (1994). A v egetation c lassificationlogic b ased o n r emote s ensing f or u se i n g lobal s cale b iogeochemical m odels, Ambio, 23, 77-‐81.17.Scepan, J. 1999. T hematic V alidation o f H igh-‐Resolution G lobal L and-‐Cover D ataSets, P hotogrammetric E ngineering a nd R emote S ensing, 65, 1051-‐1060.18.Schaaf, C.B., G ao, F., S trahler, A. H., L ucht, W., L i, X., T sang, T., S trugnell, N. C.,Zhang, X., J in, Y., M uller, J. P., L ewis, P., B arnsley, M., H obson, P., D isney, M.,Roberts, G., D underdale, M., D oll, C., d’Entremont, R. P., H u, B., L iang, S., P rivette, J.L., & R oy, D. (2002). F irst o perational B RDF, a lbedo n adir r eflectance p roducts from M ODIS. R emote S ensing o f E nvironment, 83, 135-‐148.19.Wan, Z. M., Z hang, Y. L., Z hang, Q. C., a nd L i, Z. L. (2002). V alidation o f t he l and-‐surface t emperature p roducts r etrieved f rom T erra M oderate R esolutionImaging S pectroradiometer d ata. R emote S ensing o f E nvironment, 83, 163-‐180.。

点过程建模及应用课程教学大纲CourseOutline

课程教学大纲（course syllabus）
*学习目标 (Learning Outcomes)
1．了解事件序列建模问题的特点与挑战，及其应用场景 2. 点过程的基本问题定义、基本模型，概览性的认识 2．了解面向点过程学习的前沿机器学习技术与研究现状 3．完成点过程建模或应用编程作业，对相关编程有初步训练 4．完成点过程前沿研究论文解读和相关口头报告与讨论
新趋势，进行点过程建模与学习的基本方法与及相关应用的讲解。本课程将涵盖传统统计
学领域的点过程方法（基于最大后验概率为主）与近年来结合深度学习的模型与算法（以
最大后验概率结合对抗生成网络等技术为主）。此外，本课程也将提供一定的编程类任务，
培养学生动手能力与解决开放性问题的综合能力。
*课程简介（Description）
方法，面向异步、高维事件序列的建模与学习，不仅具有普遍的应用价值，也将促进机器
学习等学科的发展。作为随机过程的一个重要方法，点过程对连续时间域事件序列具有丰
*课程简介（Description）
富的刻画能力，并具有坚实的理论基础作为支撑。本课程提供点过程的入门基础引导，特别是结合机器学习领域的相关建模思想、技术与最
授课教师（Instl
(Course Webpage)
在包括网购、发帖、机器故障等很多场合，每时每刻产生着大量事件数据。这些数据是人
们理解、预测乃至调控各类事物的重要途径。相比于回归、分类或者时间序列分析等经典
点过程建模及应用课程教学大纲
Course Outline
课程基本信息（Course Information）
课程代码（Course Code） CSXXX
*学时（Credit Hours）
48

基于聚类与自适应ALGBM_的预测模型研究

第 22卷第 3期2023年 3月Vol.22 No.3Mar.2023软件导刊Software Guide基于聚类与自适应ALGBM的预测模型研究廖雪超1，2，马亚文1，2（1.武汉科技大学计算机科学与技术学院；2.智能信息处理与实时工业系统重点实验室，湖北武汉 430065）摘要：建筑能耗预测在建筑能源管理、节能和故障诊断等方面发挥着重要作用，而建筑能耗数据之间存在非线性和离群值点，导致能耗预测精度降低。

为解决以上问题，提出基于特征提取、聚类和改进LGBM的MRGALnet建筑能耗预测模型。

首先通过MI+RFE二次特征选择算法筛选出对建筑能耗影响最大的特征子集，然后利用GMM高斯混合模型算法将能耗特性相似的建筑进行归类，并采用LGBM模型对每个聚类的能耗数据进行预测，进一步设计自适应损失函数以改进LGBM的预测性能。

通过对比实验可知，MI+RFE特征选择算法能有效去除冗余特征，GMM聚类方法则能对原始数据进行合理的聚类划分，而ALGBM模型可根据不同聚类的能耗数据自适应地确定损失函数超参数，以提高模型预测性能，综合以上算法的MRGALnet模型能够进一步提升预测精度和收敛速度。

关键词：建筑能耗预测；特征选择；聚类；轻量级梯度提升机；自适应损失函数DOI：10.11907/rjdk.222471开放科学（资源服务）标识码（OSID）：中图分类号：TP183 文献标识码：A文章编号：1672-7800（2023）003-0010-08Research on Predictive Model Based on Clustering and Adaptive ALGBMLIAO Xue-chao1，2， MA Ya-wen1，2（1.College of Computer Science and Technology， Wuhan University of Science and Technology；2.Key Laboratory of Intelligent Information Processing and Real-time Industrial Systems， Wuhan 430065， China）Abstract：Building energy consumption prediction plays an important role in building energy management， energy conservation and fault di⁃agnosis. However， there are nonlinear and outlier points among building energy consumption data， which leads to the decrease of energy con⁃sumption prediction accuracy. To solve the above problems， the MRGALnet building energy consumption prediction model based on feature ex⁃traction， clustering and improved LGBM is proposed. Firstly， the subsets of features that have the greatest impact on building energy consump⁃tion are selected through MI+RFE secondary feature selection algorithm. Secondly， building data with similar energy consumption characteris⁃tics are grouped by Gaussian mixture clustering algorithm. Thirdly， energy consumption data for each cluster are predicted by LGBM. Furter more， an adaptive loss function is designed to improve the prediction performance of LGBM. Through comparative experimental analysis， it can be seen that MI+RFE feature selection algorithm can effectively remove redundant features， GMM can reasonably cluster the original da⁃ta， and ALGBM model can adaptively determine the hyperparameters of the loss function according to the energy consumption data of different clustering， so as to improve the model prediction performance. The MRGALnet model combined with the above algorithms is optimal in terms of prediction accuracy and convergence speed. The MRGALnet model integrating the above algorithms can further improve the prediction accu⁃racy and convergence speed.Key Words：building energy consumption prediction； feature selection； clustering； light gradient boosting machine； adaptive loss function0 引言随着时代的发展，近年来能源消耗量持续增长，能源问题已成为一个全球性问题。

人工神经网络及应用智慧树知到课后章节答案2023年下长安大学

人工神经网络及应用智慧树知到课后章节答案2023年下长安大学长安大学第一章测试1.Synapse is the place where neurons connect in function. It is composed ofpresynaptic membrane, synaptic space and postsynaptic membrane.（）A:对 B:错答案:对2.Biological neurons can be divided into sensory neurons, motor neurons and（）according to their functions.A:multipolar neurons B:interneuronsC:Pseudo unipolar neural networks D:bipolar neurons答案:interneurons3.Neurons and glial cells are the two major parts of the nervous system. （）A:错 B:对答案:对4.Neurons are highly polarized cells, which are mainly composed of two parts:the cell body and the synapse. （）A:错 B:对答案:对5.The human brain is an important part of the nervous system, which containsmore than 86 billion neurons. It is the central information processingorganization of human beings. （）A:对 B:错答案:对第二章测试1.In 1989, Mead, the father of VLSI, published his monograph "( )", in which agenetic neural network model based on evolutionary system theory wasproposed.A:Learning MachinesB:Journal Neural NetworksC:Analog VLSI and Neural SystemsD:Perceptrons: An Introduction to Computational Geometry答案:Analog VLSI and Neural Systems2.In 1989, Yann Lecun proposed convolutional neural network and applied itto image processing, which should be the earliest application field of deeplearning algorithm. （）A:对 B:错答案:对3.In 1954, Eccles, a neurophysiologist at the University of Melbourne,summarized the principle of Dale, a British physiologist, that "each neuronsecretes only one kind of transmitter ".（）A:错 B:对答案:对4.In 1972, Professor Kohonen of Finland proposed a self-organizing featuremap (SOFM) neural network model. （）A:对 B:错答案:对5.Prediction and evaluation is an activity of scientific calculation andevaluation of some characteristics and development status of things orevents in the future according to the known information of objective objects.（）A:对 B:错答案:对第三章测试1.The function of transfer function in neurons is to get a new mapping outputof summer according to the specified function relationship, and thencompletes the training of artificial neural network. （）A:对 B:错答案:对2.The determinant changes sign when two rows (or two columns) areexchanged. The value of determinant is zero when two rows (or two columns) are same. （）A:对 B:错答案:对3.There are two kinds of phenomena in the objective world. The first is thephenomenon that will happen under certain conditions, which is calledinevitable phenomenon. The second kind is the phenomenon that may ormay not happen under certain conditions, which is called randomphenomenon. （）A:错 B:对答案:对4.Logarithmic S-type transfer function, namely Sigmoid function, is also calledS-shaped growth curve in biology. （）A:错 B:对答案:对5.Rectified linear unit (ReLU), similar to the slope function in mathematics, isthe most commonly used transfer function of artificial neural network. （）A:错 B:对答案:对第四章测试1.The perceptron learning algorithm is driven by misclassification, so thestochastic gradient descent method is used to optimize the loss function. （）A:misclassification B:maximum C:minimumD:correct答案:misclassification2.Perceptron is a single-layer neural network, or neuron, which is the smallestunit of neural network. （）A:错 B:对答案:对3.When the perceptron is learning, each sample will be input into the neuronas a stimulus. The input signal is the feature of each sample, and the expected output is the category of the sample. When the output is different from the category, we can adjust the synaptic weight and bias value until the output of each sample is the same as the category. （）A:对 B:错答案:对4.If the symmetric hard limit function is selected for the transfer function, theoutput can be expressed as . If the inner product of the row vector and the input vector in the weight matrix is greater than or equal to -b, the output is 1, otherwise the output is -1. （）A:错 B:对答案:对5.The basic idea of perceptron learning algorithm is to input samples into thenetwork step by step, and adjust the weight matrix of the network according to the difference between the output result and the ideal output, that is tosolve the optimization problem of loss function L(w,b). （）A:错 B:对答案:对第五章测试1.The output of BP neural network is （）of neural network.A:the output of the last layer B:the input of the last layerC:the output of the second layer D:the input of the second layer答案:the output of the last layer2.BP neural network has become one of the most representative algorithms inthe field of artificial intelligence. It has been widely used in signal processing, pattern recognition, machine control (expert system, data compression) and other fields. （）A:对 B:错答案:对3.In 1974, Paul Werbos of the natural science foundation of the United Statesfirst proposed the use of error back propagation algorithm to train artificialneural networks in his doctoral dissertation of Harvard University, anddeeply analyzed the possibility of applying it to neural networks, effectivelysolving the XOR loop problem that single sensor cannot handle. （）A:对 B:错答案:对4.In the standard BP neural network algorithm and momentum BP algorithm,the learning rate is a constant that remains constant throughout the training process, and the performance of the learning algorithm is very sensitive tothe selection of the learning rate. （）答案:对5.L-M algorithm is mainly proposed for super large scale neural network, andit is very effective in practical application. （）A:对 B:错答案:错第六章测试1.RBF neural network is a novel and effective feedforward neural network,which has the best local approximation and global optimal performance. （）A:对 B:错答案:对2.At present, RBF neural network has been successfully applied in nonlinearfunction approximation, time series analysis, data classification, patternrecognition, information processing, image processing, system modeling,control and fault diagnosis. （）A:对 B:错答案:对3.The basic idea of RBF neural network is to use radial basis function as the"basis" of hidden layer hidden unit to form hidden layer space, and hiddenlayer transforms input vector. The input data transformation of lowdimensional space is mapped into high-dimensional space, so that theproblem of linear separability in low-dimensional space can be realized inhigh-dimensional space. （）答案:对4.For the learning algorithm of RBF neural network, the key problem is todetermine the center parameters of the output layer node reasonably. （）A:对 B:错答案:错5.The method of selecting the center of RBF neural network by self-organizinglearning is to select the center of RBF neural network by k-means clustering method, which belongs to supervised learning method. （）A:错 B:对答案:错第七章测试1.In terms of algorithm, ADALINE neural network adopts W-H learning rule,also known as the least mean square (LMS) algorithm. It is developed fromthe perceptron algorithm, and its convergence speed and accuracy have been greatly improved. （）A:错 B:对答案:对2.ADALINE neural network has simple structure and multi-layer structure. It isflexible in practical application and widely used in signal processing, system identification, pattern recognition and intelligent control. （）A:对 B:错答案:对3.When there are multiple ADALINE in the network, the adaptive linear neuralnetwork is also called Madaline which means many Adaline neural networks.（）A:对 B:错答案:对4.The algorithm used in single-layer ADALINE network is LMS algorithm,which is similar to the algorithm of perceptron, and also belongs tosupervised learning algorithm. （）A:对 B:错答案:对5.In practical application, the inverse of the correlation matrix and thecorrelation coefficient are not easy to obtain, so the approximate steepestdescent method is needed in the algorithm design. The core idea is that theactual mean square error of the network is replaced by the mean squareerror of the k-th iteration.（）A:错 B:对答案:对第八章测试1.Hopfield neural network is a kind of neural network which combines storagesystem and binary system. It not only provides a model to simulate humanmemory, but also guarantees the convergence to （）.A:local minimum B:local maximumC:minimumD:maximum答案:local minimum2.At present, researchers have successfully applied Hopfield neural network tosolve the traveling salesman problem (TSP), which is the most representative of optimization combinatorial problems. （）A:错 B:对答案:对3.In 1982, American scientist John Joseph Hopfield put forward a kind offeedback neural network "Hopfield neural network" in his paper NeuralNetworks and Physical Systems with Emergent Collective ComputationalAbilities. （）A:对 B:错答案:对4.Under the excitation of input x, DHNN enters a dynamic change process, untilthe state of each neuron is no longer changed, it reaches a stable state. This process is equivalent to the process of network learning and memory, and the final output of the network is the value of each neuron in the stable state.（）A:错 B:对答案:对5.The order in which neurons adjust their states is not unique. It can beconsidered that a certain order can be specified or selected randomly. The process of neuron state adjustment includes three situations: from 0 to 1, and1 to 0 and unchanged. （）A:错 B:对答案:对第九章测试pared with GPU, CPU has higher processing speed, and has significantadvantages in processing repetitive tasks. （）A:对 B:错答案:错2.At present, DCNN has become one of the core algorithms in the field of imagerecognition, but it is unstable when there is a small amount of learning data.（）A:对 B:错答案:错3.In the field of target detection and classification, the task of the last layer ofneural network is to classify. （）A:对 B:错答案:对4.In AlexNet, there are 650000 neurons with more than 600000 parametersdistributed in five convolution layers and three fully connected layers andSoftmax layers with 1000 categories. （）A:对 B:错答案:错5.VGGNet is composed of two parts: the convolution layer and the fullconnection layer, which can be regarded as the deepened version of AlexNet.（）A:错 B:对答案:对第十章测试1.The essence of the optimization process of D and G is to find the（）.A:maximum B:minimax C:local maximaD:minimum答案:minimax2.In the artificial neural network, the quality of modeling will directly affect theperformance of the generative model, but a small amount of prior knowledge is needed for the actual case modeling.（）A:对 B:错答案:错3. A GAN mainly includes a generator G and a discriminator D. （）A:对 B:错答案:对4.Because the generative adversarial network does not need to distinguish thelower bound and approximate inference, it avoids the partition functioncalculation problem caused by the traditional repeated application of Markov chain learning mechanism, and improves the network efficiency. （）A:对 B:错答案:对5.From the perspective of artificial intelligence, GAN uses neural network toguide neural network, and the idea is very strange. （）A:对 B:错答案:对第十一章测试1.The characteristic of Elman neural network is that the output of the hiddenlayer is delayed and stored by the feedback layer, and the feedback isconnected to the input of the hidden layer, which has the function ofinformation storage. （）A:对 B:错答案:对2.In Elman network, the transfer function of feedback layer is nonlinearfunction, and the transfer function of output layer islinear function.（）A:对 B:错答案:对3.The feedback layer is used to memorize the output value of the previous timeof the hidden layer unit and return it to the input. Therefore, Elman neuralnetwork has dynamic memory function.（）A:对 B:错答案:对4.The neurons in the hidden layer of Elman network adopt the tangent S-typetransfer function, while the output layer adopts the linear transfer function. If there are enough neurons in the feedback layer, the combination of thesetransfer functions can make Elman neural network approach any functionwith arbitrary precision in finite time.（）A:对 B:错答案:对5.Elman neural network is a kind of dynamic recurrent network, which can bedivided into full feedback and partial feedback. In the partial recurrentnetwork, the feedforward connection weight can be modified, and thefeedback connection is composed of a group of feedback units, and theconnection weight cannot be modified. （）A:错 B:对答案:对第十二章测试1.The loss function of AdaBoost algorithm is（）.A:exponential functionB:nonlinear function C:linear functionD:logarithmic function答案:exponential function2.Boosting algorithm is the general name of a class of algorithms. Theircommon ground is to construct a strong classifier by using a group of weakclassifiers. Weak classifier mainly refers to the classifier whose predictionaccuracy is not high and far below the ideal classification effect. Strongclassifier mainly refers to the classifier with high prediction accuracy. （）A:错 B:对答案:对3.Among the many improved boosting algorithms, the most successful one isthe AdaBoost (adaptive boosting) algorithm proposed by Yoav Freund ofUniversity of California San Diego and Robert Schapire of PrincetonUniversity in 1996. （）A:错 B:对答案:对4.The most basic property of AdaBoost is that it reduces the training errorcontinuously in the learning process, that is, the classification error rate onthe training data set until each weak classifier is combined into the final ideal classifier. （）A:错 B:对答案:对5.The main purpose of adding regularization term into the formula ofcalculating strong classifier is to prevent the over fitting of AdaBoostalgorithm, which is usually called step size in algorithm. （）A:错 B:对答案:对第十三章测试1.The core layer of SOFM neural network is（）.A:input layer B:hidden layerC:output layer D:competition layer答案:competition layer2.In order to divide the input patterns into several classes, the distancebetween input pattern vectors should be measured according to thesimilarity. （）are usually used.A:Euclidean distance method B:Cosine methodC:Sine method D:Euclidean distance method and cosine method答案:Euclidean distance method and cosine method3.SOFM neural networks are different from other artificial neural networks inthat they adopt competitive learning rather than backward propagationerror correction learning method similar to gradient descent, and in a sense, they use neighborhood functions to preserve topological properties of input space. （）A:对 B:错答案:对4.For SOFM neural network, the competitive transfer function (CTF) responseis 0 for the winning neurons, and 1 for other neurons.（）A:错 B:对答案:错5.When the input pattern to the network does not belong to any pattern in thenetwork training samples, SOFM neural network can only classify it into the closest mode. （）A:对 B:错答案:对第十四章测试1.The neural network toolbox contains（）module libraries.A:three B:sixC:five D:four答案:five2.The "netprod" in the network input module can be used for（）.A:dot multiplication B:dot divisionC:addition or subtractionD:dot multiplication or dot division答案:dot multiplication or dot division3.The "dotrod" in the weight setting module is a normal dot product weightfunction.（）A:错 B:对答案:错4.The mathematical model of single neuron is y=f(wx+b).（）A:错 B:对答案:对5.The neuron model can be divided into three parts: input module, transferfunction and output module. （）A:对 B:错答案:对第十五章测试1.In large-scale system software design, we need to consider the logicalstructure and physical structure of software architecture. （）A:对 B:错答案:对2.The menu property bar has "label" and "tag". The label is equivalent to thetag value of the menu item, and the tag is the name of the menu display.（）A:对 B:错答案:错3.It is necessary to determine the structure and parameters of the neuralnetwork, including the number of hidden layers, the number of neurons inthe hidden layer and the training function.（）A:对 B:错答案:对4.The description of the property "tooltipstring" is the prompt that appearswhen the mouse is over the object. （）A:对 B:错答案:对5.The description of the property "string" is: the text displayed on the object.（）A:错 B:对答案:对第十六章测试1.The description of the parameter "validator" of the wx.TextCtrl class is: the（）.A:size of controlB:style of control C:validator of controlD:position of control答案:validator of control2.The description of the parameter "defaultDir" of class wx.FileDialog is: （）.A:open the file B:default file nameC:default path D:save the file答案:default path3.In the design of artificial neural network software based on wxPython,creating GUI means building a framework in which various controls can beadded to complete the design of software functions. （）A:对 B:错答案:对4.When the window event occurs, the main event loop will respond and assignthe appropriate event handler to the window event. （）A:对 B:错答案:对5.From the user's point of view, the wxPython program is idle for a large partof the time, but when the user or the internal action of the system causes the event, and then the event will drive the wxPython program to produce the corresponding action.（）A:对 B:错答案:对。

pisa2006科学试题

Document: ReleasedPISAItems_Science.docPISA RELEASED ITEMS - SCIENCEDecember 2006Table of ContentsS126: Biodiversity (3)S127: Buses (6)S128: Cloning (8)S129: Daylight (11)S195: Semmelweis’ Diary (16)S210: Climate Change (22)S212: Flies (24)S251: Calf Clones (28)S253: Ozone (31)S307: Corn (37)S409: Fit for Drinking (40)S414: Tooth Decay (45)S420: Hot Work (48)S423: Mousepox (50)S433: Stickleback Behaviour (53)S439: Tobacco Smoking (59)S441: Starlight (63)S448: Ultrasound (64)S470: Lip Gloss (67)S472: Evolution (69)S505: Bread Dough (72)S507: Transit of Venus (76)S515: Health Risk? (79)S516: Catalytic Converter (82)S526: Major Surgery (86)S529: Wind Farms (90)Source Publications for Released Items (94)S126: BiodiversityBiodiversity Text 1Read the following newspaper article and answer the questions which follow.BIODIVERSITY IS THE KEY TO MANAGING ENVIRONMENTAn ecosystem that retains a high biodiversity (that is, a wide variety of living things) is much more likely to adapt to human-caused environment change than is one that has little. Consider the two food webs shown in the diagram. The arrows point from the organism that gets eaten to the one that eats it. These food webs are highly5simplified compared with food webs in real ecosystems, but they still illustrate a key difference between more diverse and less diverse ecosystems. Food web B represents a situation with very low biodiversity, where at some levels the food path involves only a single type of organism. Food web A represents a more diverse ecosystem with, as a result, many more alternative feeding pathways. 10Generally, loss of biodiversity should be regarded seriously, not only because the organisms that have become extinct represent a big loss for both ethical and utilitarian (useful benefit) reasons, but also because the organisms that remain have become more vulnerable (exposed) to extinction in the future.Source: Adapted from Steve Malcolm: ‘Biodiversity is the key to managing environment’, The Age , 16 August 1994.FOOD WEB A FOOD WEB BEucalypt Beetle Spider Lizard SnakeWattle Tea TreeLeaf HopperButterfly LarvaeParasitic Wasp HoneyeaterRobinButcher BirdNative CatNative CatButcher BirdSnakeLizard RobinParasitic WaspLeaf HopperWattleQuestion 3: BIODIVERSITY S126Q03 In lines 9 and 10 it is stated that “Food web A represents a more diverse ecosystem with, as a result, many more alternative feeding pathways.”Look at FOOD WEB A. Only two animals in this food web have three direct (immediate) food sources. Which two animals are they?A Native Cat and Parasitic WaspB Native Cat and Butcher BirdC Parasitic Wasp and Leaf HopperD Parasitic Wasp and SpiderE Native Cat and HoneyeaterBIODIVERSITY SCORING 3QUESTION INTENT: Process: Demonstrating knowledge and understandingTheme: EcosystemsArea: Science in life and healthFull creditCode 1: A. Native Cat and Parasitic WaspNo creditCode 0: Other responses.Code 9: Missing.Question 4: BIODIVERSITY S126Q04 Food webs A and B are in different locations. Imagine if Leaf Hoppers died out in both locations. Which one of these is the best prediction and explanation for the effect this would have on the food webs?A The effect would be greater in food web A because the Parasitic Wasp has onlyone food source in web A.B The effect would be greater in food web A because the Parasitic Wasp hasseveral food sources in web A.C The effect would be greater in food web B because the Parasitic Wasp has onlyone food source in web B.D The effect would be greater in food web B because the Parasitic Wasp hasseveral food sources in web B.BIODIVERSITY SCORING 4QUESTION INTENT: Process: Drawing/evaluating conclusionsTheme: BiodiversityArea: Science in life and healthFull creditCode 1: C. The effect would be greater in food web B because the Parasitic Wasp has only one food source in web B.No creditCode 0: Other responses.Code 9: Missing.S127: BusesQuestion 1: BUSES S127Q01 A bus is driving along a straight stretch of road. The bus driver, named Ray, has a cup of water resting on the dashboard:1 2waterdriving directionSuddenly Ray has to slam on the brakes.What is most likely to happen to the water in the cup?A The water will stay horizontal.B The water will spill over side 1.C The water will spill over side 2.D The water will spill but you cannot tell if it will spill at side 1 or side 2.BUSES SCORING 1QUESTION INTENT: Process: Demonstrating knowledge and understandingTheme: Forces and movementArea: Science in technologiesFull creditCode 1: C. The water will spill over side 2.No creditCode 0: Other responses.Code 9: Missing.Question 4: BUSES S127Q04-0189 Ray’s bus is, like most buses, powered by a petrol engine. These buses contribute to environmental pollution.Some cities have trolley buses: they are powered by an electric engine. The voltage needed for such an electric engine is provided by overhead lines (like electric trains). The electricity is supplied by a power station using fossil fuels.Supporters for the use of trolley buses in a city say that these buses don’t contribute to environmental pollution.Are these supporters right? Explain your answer. .................................................... ................................................................................................................................... ................................................................................................................................... ...................................................................................................................................BUSES SCORING 4QUESTION INTENT: Process: Demonstrating knowledge and understandingTheme: Energy transformationsArea: Science in Earth and environmentFull creditCode1: Gives an answer in which it is stated that the power station also contributes to environmental pollution:• No, because the power station causes environmental pollution as well.• Yes, but this is only true for the city itself; the power station however causesenvironmental pollution.No creditCode 0: No or yes, without a correct explanation.Code 8: Off task.Code 9: Missing.Example responsesCode 1:• Yes and No. The buses don’t pollute the city which is good, but the power stationdoes pollute and that’s not very good.• The buses do contribute to the environmental pollution by using fossil fuels butthey’re not as harmful as normal buses with all their gases. [Note: This answercan be given the benefit of the doubt.]Code 0:• Well they have no outlet so no harmful smoke goes into the air which candamage the O-zone layer, and having electricity created by fossil fuels is alsomore environmental friendly.• Yes, they are. Because electricity isn’t harmful for the environment we only useup our Earth’s gas.S128: CloningRead the newspaper article and answer the questions that follow.Question 1: CLONING S128Q01 Which sheep is Dolly identical to?A Sheep 1B Sheep 2C Sheep 3D Dolly’s fatherCLONING SCORING 1Full creditCode 1: A. Sheep 1No creditCode 0: Other responses.Code 9: Missing.Question 2: CLONING S128Q02 In line 14 the part of the udder that was used is described as “a very small piece”. From the article text you can work out what is meant by “a very small piece”.That “very small piece” isA a cell.B a gene.C a cell nucleus.D a chromosome.CLONING SCORING 2Full creditCode 1: A. a cell.No creditCode 0: Other responses.Code 9: Missing.Question 3: CLONING S128Q03In the last sentence of the article it is stated that many governments have alreadydecided to forbid cloning of people by law.Two possible reasons for this decision are mentioned below.Are these reasons scientific reasons?Circle either “Yes” or “No” for each.Reason: Scientific? Cloned people could be more sensitive to certain diseases thanYes / Nonormal people.People should not take over the role of a Creator. Yes / NoCLONING SCORING 3Full creditCode 1: Yes, No, in that order.No creditCode 0: Other responses.Code 9: Missing.S129: DaylightRead the following information and answer the questions that follow. DAYLIGHT ON 22 JUNE 2002Today, as the Northern Hemisphere celebrates its longest day, Australians will experience their shortest.In Melbourne*, Australia, the Sun will rise at 7:36 am and set at 5:08 pm, giving nine hours and 32 minutes of daylight. Compare today to the year’s longest day in the Southern Hemisphere, expected on 22 December, when the Sun will rise at 5:55 am and set at 8:42 pm, giving 14 hours and 47 minutes of daylight.The President of the Astronomical Society, Mr Perry Vlahos, said the existence of changing seasons in the Northern and Southern Hemispheres was linked to the Earth’s 23-degree tilt.*Melbourne is a city in Australia at a latitude of about 38 degrees South of the equator.Question 1: DAYLIGHT S129Q01 Which statement explains why daylight and darkness occur on Earth?A The Earth rotates on its axis.B The Sun rotates on its axis.C The Earth’s axis is tilted.D The Earth revolves around the Sun.DAYLIGHT SCORING 1Full creditCode 1: A. The Earth rotates on its axis.No creditCode 0: Other responses.Code 9: Missing.Question 2: DAYLIGHT S129Q02 - 01 02 03 04 11 12 13 21 99 In the Figure light rays from the Sun are shown shining on the Earth.Suppose it is the shortest day in Melbourne.Show the Earth’s axis, the Northern Hemisphere, the Southern Hemisphere and the Equator on the Figure. Label all parts of your answer.DAYLIGHT SCORING 2Note: the important features when marking this question are:1. The Earth’s axis is drawn tilted towards the Sun within the range 10° and 45° from vertical for credit: refer to the following diagram:Outside of 10° and 45° to vertical range: no credit.2. The presence or absence of clearly labelled Northern and Southern Hemispheres, or one Hemisphere only labelled, the other implied.3. The equator is drawn at a tilt towards the Sun within the range 10° and 45° above horizontal for credit: refer to the following diagram:CREDIT FOR AXIS10O 23O 45OFigure: light rays from SunThe equator may be drawn as an elliptical line or straight line.Outside of 10° and 45° to horizontal range: no credit.Full creditCode 21: Diagram with Equator tilted towards the Sun at an angle between 10° and45° and Earth’s axis tilted towards the Sun within the range 10° and 45°from vertical, and the Northern and or Southern Hemispheres correctlylabelled (or one only labelled, the other implied).Partial creditCode 11: Angle of tilt of axis between 10° and 45°, Northern and / or SouthernHemispheres correctly labelled (or one only labelled, the other implied), butangle of tilt of Equator not between 10° and 45°; or Equator missing.CREDIT FOR EQUATOR 10O23O45ONS A EquatorAxis N Equator Axis N NAxisS S EquatorCode 12: Angle of tilt of Equator between 10° and 45°, Northern and / or SouthernHemispheres correctly labelled (or one only labelled, the other implied), butangle of tilt of axis not between 10° and 45°; or axis missing.Code 13: Angle of tilt of Equator between 10° and 45°, and angle of tilt of axisbetween 10° and 45°, but Northern and Southern Hemispheres notcorrectly labelled (or one only labelled, the other implied, or both missing).No creditCode 01: Northern and or Southern Hemispheres correctly labelled (or one only, theother implied) is the only correct feature.Code 02: Angle of tilt of Equator between 10° and 45° is the only correct feature.NS AxisEquator N S Axis Equator N Axis Equator Axis EquatorN SEquatorCode 03: Angle of tilt of axis between 10° and 45° is the only correct feature.AxisCode 04: No features are correct, or other responses.SNCode 99: Missing.S195: Semmelweis’ DiarySemmelweis’ Diary Text 1‘July 1846. Next week I will take up a position as “Herr Doktor” at the First Ward of the maternity clinic of the Vienna General Hospital. I was frightened when I heard about the percentage of patients who die in this clinic. This month not less than 36 of the 208 mothers died there, all from puerperal fever. Giving birth to a child is as dangerous as first-degree pneumonia.’These lines from the diary ofIgnaz Semmelweis (1818-1865)illustrate the devastating effects of puerperal fever, a contagious disease that killed many women after childbirth. Semmelweiscollected data about the numberof deaths from puerperal fever in both the First and the SecondWards (see diagram).Physicians, among them Semmelweis, were completely in the dark about the cause of puerperal fever. Semmelweis’ diary again:‘December 1846. Why do so many women die from this fever after giving birth without any problems? For centuries science has told us that it is an invisible epidemic that kills mothers. Causes may be changes in the air or some extraterrestrial influence or a movement of the earth itself, an earthquake.’Nowadays not many people would consider extraterrestrial influence or anearthquake as possible causes of fever. But in the time Semmelweis lived, many people, even scientists, did! We now know it has to do with hygienic conditions. Semmelweis knew that it was unlikely that fever could be caused by extraterrestrial influence or an earthquake. He pointed at the data he collected (see diagram) and used this to try to persuade his colleagues.Diagram184118421843184418451846Year15105Number of Deaths First WardSecondWardNumber of Deaths per 100 deliveries from puerperal feverQuestion 2: SEMMELWEIS’ DIARY S195Q02- 01 02 03 04 11 12 13 21 99 Suppose you were Semmelweis. Give a reason (based on the data Semmelweis collected) why puerperal fever is unlikely to be caused by earthquakes. ................................................................................................................................... ................................................................................................................................... ................................................................................................................................... ...................................................................................................................................SEMMELWEIS’ DIARY SCORING 2QUESTION INTENT: Process: Drawing/evaluating conclusionsTheme: Human biologyArea: Science in life and healthFull creditCode 21: Refers to the difference between the number of deaths (per 100 deliveries) in both wards.• Due to the fact that the first ward had a high rate of women dying compared towomen in the second ward, obviously shows that it had nothing to do withearthquakes.• Not as many people died in ward 2 so an earthquake couldn’t have occurredwithout causing the same number of deaths in each ward.•Because the second ward isn’t as high, maybe it had something to do with ward 1.• It is unlikely that earthquakes cause the fever since death rates are so differentfor the two wards.Partial creditCode 11: Refers to the fact that earthquakes don’t occur frequently.• It would be unlikely to be caused by earthquakes because earthquakes wouldn’thappen all the time.Code 12: Refers to the fact that earthquakes also influence people outside the wards.• If there were an earthquake, women from outside the hospital would have gotpuerperal fever as well.• If an earthquake were the reason, the whole world would get puerperal fevereach time an earthquake occurs (not only the wards 1 and 2).Code 13: Refers to the thought that when earthquakes occur, men don’t get puerperal fever.• If a man were in the hospital and an earthquake came, he didn’t get puerperalfever, so earthquakes cannot be the cause.• Because girls get it and not men.No creditCode 01: States (only) that earthquakes cannot cause the fever.• An earthquake cannot influence a person or make him sick.• A little shaking cannot be dangerous.Code 02: States (only) that the fever must have another cause (right or wrong).• Earthquakes do not let out poison gases. They are caused by the plates of theEarth folding and faulting into each other.• Because they have nothing to do with each other and it is just superstition.• An earthquake doesn’t have any influence on the pregnancy. The reason wasthat the doctors were not specialised enough.Code 03: Answers that are combinations of Codes 01 and 02.• Puerperal fever is unlikely to be caused by earthquakes as many women dieafter giving birth without any problems. Science has told us that it is an invisibleepidemic that kills mothers.• The death is caused by bacteria and the earthquakes cannot influence them. Code 04: Other responses.• I think it was a big earthquake that shook a lot.• In 1843 the deaths decreased at ward 1 and less so at ward 2.• Because there aren’t any earthquakes by the wards and they still got it. [Note:The assumption that there were no earthquakes at that time isn’t correct.] Code 99: Missing.Semmelweis’ Diary Text 2Part of the research in the hospital was dissection. The body of a deceased person was cut open to find a cause of death. Semmelweis recorded that the students working on the First ward usually took part in dissections on women who died the previous day, before they examined women who had just given birth. They did not pay much attention to cleaning themselves after the dissections. Some were even proud of the fact that you could tell by their smell that they had been working in the mortuary, as this showed how industrious they were!One of Semmelweis’ friends died after having cut himself during such a dissection. Dissection of his body showed he had the same symptoms as mothers who died from puerperal fever. This gave Semmelweis a new idea.Question 4: SEMMELWEIS’ DIARY S195Q04 Semmelweis’ new idea had to do with the high percentage of women dying in the maternity wards and the students’ behaviour.What was this idea?A Having students clean themselves after dissections should lead to a decrease ofpuerperal fever.B Students should not take part in dissections because they may cut themselves.C Students smell because they do not clean themselves after a dissection.D Students want to show that they are industrious, which makes them carelesswhen they examine the women.SEMMELWEIS’ DIARY SCORING 4QUESTION INTENT: Process: Recognising questionsTheme: Human biologyArea: Science in life and healthFull creditCode 1: A. Having students clean themselves after dissections should lead to a decrease of puerperal fever.No creditCode 0: Other responses.Code 9: Missing.Question 5: SEMMELWEIS’ DIARY S195Q05-01 02 11 12 13 14 15 99 Semmelweis succeeded in his attempts to reduce the number of deaths due to puerperal fever. But puerperal fever even today remains a disease that is difficult to eliminate.Fevers that are difficult to cure are still a problem in hospitals. Many routine measures serve to control this problem. Among those measures are washing sheets at high temperatures.Explain why high temperature (while washing sheets) helps to reduce the risk that patients will contract a fever. ......................................................................................................................................................................................................................................................................SEMMELWEIS’ DIARY SCORING 5QUESTION INTENT: Process: Demonstrating knowledge and understandingTheme: Human biologyArea: Science in life and healthFull creditCode 11: Refers to killing of bacteria .• Because with the heat many bacteria will die.• Bacteria will not stand the high temperature.• Bacteria will be burnt by the high temperature.• Bacteria will be cooked. [Note: Although “burnt” and “cooked” are notscientifically correct, each of the last two answers as a whole can be regardedas correct.]Code 12: Refers to killing of microorganisms, germs or viruses.• Because high heat kills small organisms which cause disease.• It’s too hot for germs to live.Code 13: Refers to the removal (not killing) of bacteria.• The bacteria will be gone.• The number of bacteria will decrease.• You wash the bacteria away at high temperatures.Code 14: Refers to the removal (not killing) of microorganisms, germs or viruses.• Because you won’t have the germ on your body.Code 15: Refers to sterilisation of the sheets.• The sheets will be sterilised.No creditCode 01: Refers to killing of disease.• Because the hot water temperature kills any disease on the sheets.• The high temperature kills most of the fever on the sheets, leaving less chanceof contamination.Code 02: Other responses.• So they don’t get sick from the cold.• Well when you wash something it washes away the germs.Code 99: Missing.Question 6: SEMMELWEIS’ DIARY S195Q06 Many diseases may be cured by using antibiotics. However, the success of some antibiotics against puerperal fever has diminished in recent years.What is the reason for this?A Once produced, antibiotics gradually lose their activity.B Bacteria become resistant to antibiotics.C These antibiotics only help against puerperal fever, but not against otherdiseases.D The need for these antibiotics has been reduced because public health conditionshave improved considerably in recent years.SEMMELWEIS’ DIARY SCORING 6QUESTION INTENT: Process: Demonstrating knowledge and understandingTheme: BiodiversityArea: Science in life and healthFull creditCode 1: B. Bacteria become resistant to antibiotics.No creditCode 0: Other responses.Code 9: Missing.S210: Climate ChangeClimate Change Text 1Read the following information and answer the questions which follow.WHAT HUMAN ACTIVITIES CONTRIBUTE TO CLIMATE CHANGE?The burning of coal, oil and natural gas, as well as deforestation and variousagricultural and industrial practices, are altering the composition of the atmosphere and contributing to climate change. These human activities have led to increased concentrations of particles and greenhouse gases in the atmosphere. The relative importance of the main contributors to temperature change is shown in Figure 1. Increased concentrations of carbon dioxide and methane have a heating effect. Increased concentrations of particles have a cooling effect in two ways, labelled ‘Particles’ and ‘Particle effects on clouds’.Figure 1: Relative importance of the main contributors to change intemperature of the atmosphere.Bars extending to the right of the centre line indicate a heating effect. Bars extending to the left of the centre line indicate a cooling effect. The relative effect of ‘Particles’ and ‘Particle effects on clouds’ are quite uncertain: in each case the possible effect is somewhere in the range shown by the light grey bar.Source: adapted from /ipcc/qa/04.htmlCooling Relative ImportanceHeatingQuestion 1: CLIMATE CHANGE S210Q01-01289 Use the information in Figure 1 to develop an argument in support of reducing the emission of carbon dioxide from the human activities mentioned. ................................................................................................................................... ................................................................................................................................... ................................................................................................................................... CLIMATE CHANGE SCORING 1QUESTION INTENT: Process: CommunicatingTheme: The Earth and its place in the universeArea: Science in Earth and environmentFull creditCode 2: Carbon dioxide is the main factor causing an increase in atmospheric temperature/causing climatic change, so reducing the amount emitted willhave the greatest effect in reducing the impact of human activities.Partial creditCode 1: Carbon dioxide is causing an increase in atmospheric temperature/causing climatic change.No creditCode 0: Other responses, including that an increase in temperature will have a bad effect on the Earth.Code 8: Off task.Code 9: Missing.Example responsesCode 2:• The emission of CO2 causes significant heating to the atmosphere and thereforeshould be lessened. [Note: The term “significant” can be considered asequivalent to “most”. ]• According to figure 1 reduction in the emission of carbon dioxide is necessarybecause it considerably heats the earth. [Note: The term “considerable” can beconsidered as equivalent to “most”.]Code 1:• The burning of fossil fuel such as oil, gas and coal are contributing to the buildup of gases in the atmosphere, one of which is carbon dioxide (CO2). This gasaffects the temperature of the earth which increases causing a greenhouseeffect.Code 0:• The way that humans could help control carbon dioxide levels to drop would beby not driving a car, don’t burn coal and don’t chop down forests. [Note: Noconsideration given to the effect of carbon dioxide on temperature.]S212: FliesFlies Text 1Read the following information and answer the questions which follow.FLIESA farmer was working with dairy cattle at an agricultural experiment station. The population of flies in the barn where the cattle lived was so large that the animals’ health was affected. So the farmer sprayed the barn and the cattle with a solution of insecticide A. The insecticide killed nearly all the flies. Some time later, however,the number of flies was again large. The farmer again sprayed with the insecticide. The result was similar to that of the first spraying. Most, but not all, of the flies were killed. Again, within a short time the population of flies increased, and they were again sprayed with the insecticide. This sequence of events was repeated five times: then it became apparent that insecticide A was becoming less and less effective in killing the flies.The farmer noted that one large batch of the insecticide solution had been made and used in all the sprayings. Therefore he suggested the possibility that the insecticide solution decomposed with age.Source: Teaching About Evolution and the Nature of Science, National Academy Press, Washington, DC, 1998, p. 75.Question 1: FLIES S212Q01-01234589 The farmer’s suggestion is that the insecticide decomposed with age. Briefly explain how this suggestion could be tested. ................................................................................................................................... ................................................................................................................................... ...................................................................................................................................FLIES SCORING 1QUESTION INTENT: Process: Identifying evidenceTheme: Chemical and physical changesArea: Science in life and healthFull creditCode 5: Applies to answers in which three variables (type of flies, age ofinsecticide, and exposure) are controlled eg. Compare the results from anew batch of the insecticide with results from the old batch on two groupsof flies of the same species that have not been previously exposed to theinsecticide.。

采访地震专家英语作文

采访地震专家英语作文Interviewing an Expert on EarthquakesI had the opportunity to sit down with renowned seismologist Dr. Emily Wilkins to discuss the latest developments in earthquake research and preparedness. As the head of the Seismic Monitoring and Prediction Center, Dr. Wilkins has dedicated her career to understanding the complex mechanisms that drive these powerful natural phenomena.Our conversation began with an overview of the current state of earthquake science. Dr. Wilkins explained that while our understanding of plate tectonics and seismic activity has advanced considerably in recent decades, there is still much to be learned. "Earthquakes remain one of the most difficult natural disasters to predict with any certainty," she acknowledged. "We can identifyhigh-risk fault lines and use historical data to assess the probability of an event occurring in a given region, but the exact timing, location, and magnitude of a quake is still extremely challenging to forecast."That said, Dr. Wilkins highlighted the significant strides that have been made in earthquake early warning systems. "In the past, the first indication that a quake was imminent would be the actual shaking. Now, with a network of seismic sensors strategically placed along fault lines, we can detect the initial seismic waves and send out alerts seconds or even tens of seconds before the destructive waves arrive." This critical advance, she noted, can provide valuable time for people to take shelter, stop trains, shut off gas lines, and implement other safety protocols.I was curious to learn more about the specific technologies and methodologies employed by seismologists. Dr. Wilkins explained that a combination of traditional monitoring equipment, advanced data analysis, and innovative modeling techniques are used to study earthquakes. "The backbone of our work is a global network of seismometers – highly sensitive instruments that measure ground motion and can detect even the slightest tremors. We also utilize satellite imagery, GPS data, and other remote sensing technologies to map fault lines and measure tectonic plate movements."Perhaps most fascinating is the role that artificial intelligence and machine learning are playing in earthquake research. "Our AI systems are able to rapidly process and interpret the massive amounts of seismic data we collect, identifying patterns and trends that might escape the human eye," Dr. Wilkins remarked. "This allowsus to develop more accurate predictive models and better understand the complex variables that contribute to seismic activity."I was eager to discuss the human impact of earthquakes and what is being done to improve preparedness and resilience. Dr. Wilkins stressed the critical importance of community-based disaster planning and infrastructure reinforcement. "Earthquakes don't have to be catastrophic events if people are educated, equipped, and empowered to respond effectively. That means having robust emergency communication systems, well-stocked supplies, and buildings and bridges designed to withstand strong shaking."She pointed to recent examples where early warning systems and well-coordinated emergency responses helped minimize casualties and damage. "In Japan, the early warning system gave residents vital seconds to take cover before the shaking started, saving countless lives du ring the 2011 Tōhoku earthquake and tsunami. And we saw how investment in seismic-resistant construction paid off in places like California, where modern buildings fared much better than older structures during recent quakes."Of course, Dr. Wilkins acknowledged that vulnerability to earthquakes remains unevenly distributed, with lower-income communities often facing the greatest risks. "Addressing this disparity in disaster preparedness is a major focus for us. We'reworking to ensure that earthquake education, mitigation measures, and emergency response capabilities reach all segments of society, not just the most affluent areas."Looking to the future, I asked Dr. Wilkins about the cutting-edge research that could transform earthquake science in the years ahead. She was particularly excited about the potential of using quantum sensing technology to detect the earliest stages of fault movements. "If we can build a network of ultra-sensitive quantum accelerometers, we might be able to identify the initial microfracturing and deformation of rock layers that precedes a major quake. This could give us hours or even days of advance warning – a game-changer for earthquake prediction."The conversation then turned to the human toll of these natural disasters and the importance of prioritizing resilience and recovery. "Earthquakes don't just destroy buildings and infrastructure," Dr. Wilkins remarked solemnly. "They also take an immense psychological and emotional toll on survivors. Supporting mental health, restoring community cohesion, and helping people rebuild their lives – that's an absolutely vital part of our work that often gets overlooked."As our interview drew to a close, I was struck by Dr. Wilkins' passion, expertise, and unwavering commitment to making the world a saferplace. Earthquakes may always hold an element of unpredictability, but through continued scientific advancement, technological innovation, and a holistic approach to disaster management, she and her colleagues are steadily reducing the human suffering caused by these powerful natural phenomena.。

多序列对位排列和进化分析

计算序列的距离，建立距离矩阵
Cat Dog Rat 3 4 5 7 6 Dog
2
Dog Rat
Cat
1
2 1 4
Rat
Cow 6
通过距离矩阵建进化树
Cow
Step1. 计算序列的距离，建立距离矩阵
对位排列，去除空格
（选择替代模型）
Uncorrected “p” distance (=observed percent sequence difference)
用于描述同源序列之间的亲缘关系的远近，应用到分子进化分析中。是构建分子进化树的基础。
Gene tree
a b
A B
Species tree
c
C
We often assume that gene trees give us species trees
注意概念：Paralogy（旁系同源/并系同源）& Orthology（直系同源）
Cladogram
Taxon B
Taxon C Taxon A Taxon D
no meaning 3 1
Phylogram
6 1 1
进化树
Ultrametric tree
Taxon B Taxon B Taxon C Taxon A Taxon D
time
时间度量树
Taxon C
Taxon A Taxon D
系统发生树术语
Rooted tree vs. Unrooted tree
有根树
无 A 根树 B
C
D
two major ways to root trees:
By midpoint or distance

classify

classifyClassify: Understanding the Importance and Benefits of Categorizing DataIntroduction:In today's data-driven world, businesses and organizations are constantly faced with a massive amount of information. To make sense of this large volume of data, we often turn to classification techniques. Classification involves categorizing data into different classes or groups based on certain criteria or attributes. It is a fundamental task in data mining and machine learning and has numerous applications across various industries. This document aims to explore the significance and benefits of data classification.I. Categorizing Data:Classification is the process of assigning items or instances to predefined categories or classes based on their characteristics or features. It involves analyzing data to identify distinct patterns and assign appropriate labels accordingly. Thiscategorization enables effective organization, interpretation, and utilization of data.II. Importance of Classification:1. Improved Data Organization:Classification enhances the organization of data by grouping similar items together. This structured format simplifies data management, making it easier to access and retrieve relevant information quickly.2. Increased Efficiency in Decision-Making:By categorizing data, decision-makers can gain valuable insights and make informed decisions. Classification helps in identifying trends, patterns, and relationships, enabling businesses to understand consumer behavior, market trends, and potential risks.3. Enhanced Data Analysis:Classification serves as a crucial step in data analysis. It enables statistical analyses, predictive modeling, and other advanced techniques to be applied to specific groups or classes, facilitating accurate predictions and identifying potential outliers.III. Benefits of Classification:1. Better Customer Segmentation:By utilizing classification techniques, businesses can divide their customer base into different segments based on demographics, preferences, or purchasing patterns. This segmentation aids in targeted marketing strategies, personalized offers, and improved customer satisfaction.2. Fraud Detection:Classification plays a vital role in fraud detection and prevention. By categorizing data into legitimate and fraudulent transactions, organizations can identify suspicious activities, patterns, or anomalies promptly. This enables proactive measures to be taken, reducing financial losses and ensuring security.3. Email Filtering and Spam Detection:Classification algorithms are instrumental in email filtering and spam detection. By categorizing emails as spam or legitimate, these algorithms can automatically separate unwanted or malicious messages from important businesscommunications, saving time and reducing the risk of security breaches.4. Medical Diagnosis and Disease Prediction:Classification techniques are widely used in the healthcare industry for medical diagnosis. By training algorithms on historical patient data, medical professionals can classify patients into different disease categories, aiding accurate diagnosis and personalized treatment options.IV. Methods of Classification:1. Decision Trees:Decision trees are a popular method for classification. They use a tree-like model to classify data based on a series of if-else conditions.2. Support Vector Machines (SVM):SVM is a supervised learning algorithm that separates data into different classes by creating a hyperplane in the feature space.3. K-Nearest Neighbors (KNN):KNN is a non-parametric algorithm that classifies data based on the majority vote of its nearest neighbors.4. Naive Bayes:Naive Bayes is a probabilistic classifier based on Bayes' theorem. It assumes independence between features and calculates the probability of an instance belonging to a specific class.V. Challenges and Limitations:While classification is a powerful tool for data analysis, it does come with certain challenges and limitations. These include the need for high-quality data, the potential bias in classification models, the complexity of handling categorical variables, and scalability issues with large datasets.Conclusion:In summary, classification is a crucial process for organizing, analyzing, and making sense of vast amounts of data. It offers numerous benefits such as improved data organization, efficient decision-making, and enhanced data analysis. From customer segmentation to fraud detection and email filtering,classification techniques find applications in various domains. Understanding and leveraging the power of classification can help businesses gain a competitive edge by uncovering valuable insights and making data-driven decisions.。

maptree包的中文名字：树模型映射、剪枝和图形功能说明书

Package‘maptree’October13,2022Version1.4-8Date2022-04-03Title Mapping,Pruning,and Graphing Tree ModelsAuthor Denis White,Robert B.Gramacy<**********>Maintainer Robert B.Gramacy<**********>Depends R(>=2.14),cluster,rpartDescription Functions with example data for graphing,pruning,andmapping models from hierarchical clustering,and classiﬁcationand regression trees.License UnlimitedRepository CRANDate/Publication2022-04-0611:52:39UTCNeedsCompilation noR topics documented:clip.clust (2)clip.rpart (3)draw.clust (4)draw.tree (5)group.clust (6)group.tree (7)kgs (8)map.groups (9)map.key (10)ngon (12)oregon.bird.dist (13)s (14)oregon.border (15)oregon.env.vars (15)oregon.grid (16)twins.to.hclust (17)Index1912clip.clust clip.clust Prunes a Hierarchical Cluster TreeDescriptionReduces a hierarchical cluster tree to a smaller tree either by pruning until a given number of observation groups remain,or by pruning tree splits below a given height.Usageclip.clust(cluster,data=NULL,k=NULL,h=NULL)Argumentscluster object of class hclust or twins.data clustered dataset for hclust application.k desired number of groups.h height at which to prune for grouping.At least one of k or h must be speciﬁed;k takes precedence if both are given.DetailsUsed with draw.clust.See example.ValuePruned cluster object of class hclust.Author(s)Denis WhiteSee Alsohclust,twins.object,cutree,draw.clustExampleslibrary(cluster)data(oregon.bird.dist)draw.clust(clip.clust(agnes(oregon.bird.dist),k=6))clip.rpart3 clip.rpart Prunes an Rpart Classiﬁcation or Regression TreeDescriptionReduces a prediction tree produced by rpart to a smaller tree by specifying either a cost-complexity parameter,or a number of nodes to which to prune.Usageclip.rpart(tree,cp=NULL,best=NULL)Argumentstree object of class rpart.cp cost-complexity parameter.best number of nodes to which to prune.If both cp and best are not NULL,then cp is used.DetailsA minor enhancement of the existing prune.rpart to incorporate the parameter best as it is usedin the(now defunct)prune.tree function in the old tree package.See example.ValuePruned tree object of class rpart.Author(s)Denis WhiteSee Alsorpart,prune.rpartExampleslibrary(rpart)data(oregon.env.vars,oregon.border,oregon.grid)draw.tree(clip.rpart(rpart(oregon.env.vars),best=7),nodeinfo=TRUE,units="species",cases="cells",digits=0)group<-group.tree(clip.rpart(rpart(oregon.env.vars),best=7))names(group)<s(oregon.env.vars)map.groups(oregon.grid,group)lines(oregon.border)4draw.clustmap.key(0.05,0.65,labels=as.character(seq(6)),size=1,new=FALSE,sep=0.5,pch=19,head="node")draw.clust Graph a Hierarchical Cluster TreeDescriptionGraph a hierarchical cluster tree of class twins or hclust using colored symbols at observations.Usagedraw.clust(cluster,data=NULL,cex=par("cex"),pch=par("pch"),size=2.5*cex, col=NULL,nodeinfo=FALSE,cases="obs",new=TRUE)Argumentscluster object of class hclust or twins.data clustered dataset for hclust application.cex size of text,par parameter.pch shape of symbol at leaves,par parameter.size size in cex units of symbol at leaves.col vector of colors from hsv,rgb,etc,or if NULL,then use rainbow.nodeinfo if TRUE,add a line at each node with number of observations included in each leaf.cases label for type of observations.new if TRUE,call plot.new.DetailsAn alternative to pltree and plot.hclust.ValueThe vector of colors supplied or generated.Author(s)Denis WhiteSee Alsoagnes,diana,hclust,draw.tree,map.groupsdraw.tree5Exampleslibrary(cluster)data(oregon.bird.dist)draw.clust(clip.clust(agnes(oregon.bird.dist),k=6))draw.tree Graph a Classiﬁcation or Regression TreeDescriptionGraph a classiﬁcation or regression tree with a hierarchical tree diagram,optionally including col-ored symbols at leaves and additional info at intermediate nodes.Usagedraw.tree(tree,cex=par("cex"),pch=par("pch"),size=2.5*cex,col=NULL,nodeinfo=FALSE,units="",cases="obs",digits=getOption("digits"),print.levels=TRUE,new=TRUE)Argumentstree object of class rpart or tree.cex size of text,par parameter.pch shape of symbol at leaves,par parameter.size if size=0,draw terminal symbol at leaves else a symbol of size in cex units.col vector of colors from hsv,rgb,etc,or if NULL,then use rainbow.nodeinfo if TRUE,add a line at each node with mean value of response,number of obser-vations,and percent deviance explained(or classiﬁed correct).units label for units of mean value of response,if regression tree.cases label for type of observations.digits number of digits to round mean value of response,if regression tree.print.levels if TRUE,print levels of factors at splits,otherwise only the factor name.new if TRUE,call plot.new.DetailsAs in plot.rpart(,uniform=TRUE),each level has constant depth.Specifying nodeinfo=TRUE, shows the deviance explained or the classiﬁcation rate at each node.A split is shown,for numerical variables,as variable<>value when the cases with lower valuesgo left,or as variable><value when the cases with lower values go right.When the splitting variable is a factor,and print.levels=TRUE,the split is shown as levels=factor=levels with the cases on the left having factor levels equal to those on the left of the factor name,and corre-spondingly for the right.6group.clustValueThe vector of colors supplied or generated.Author(s)Denis WhiteSee Alsorpart,draw.clust,map.groupsExampleslibrary(rpart)data(oregon.env.vars)draw.tree(clip.rpart(rpart(oregon.env.vars),best=7),nodeinfo=TRUE,units="species",cases="cells",digits=0)group.clust Observation Groups for a Hierarchical Cluster TreeDescriptionAlternative to cutree that orders pruned groups from left to right in draw order.Usagegroup.clust(cluster,k=NULL,h=NULL)Argumentscluster object of class hclust or twins.k desired number of groups.h height at which to prune for grouping.At least one of k or h must be speciﬁed;k takes precedence if both are given. DetailsNormally used with map.groups.See example.ValueVector of pruned cluster membershipAuthor(s)Denis Whitegroup.tree7See Alsohclust,twins.object,cutree,map.groupsExamplesdata(oregon.bird.dist,oregon.grid)group<-group.clust(hclust(dist(oregon.bird.dist)),k=6)names(group)<s(oregon.bird.dist)map.groups(oregon.grid,group)group.tree Observation Groups for Classiﬁcation or Regression TreeDescriptionAlternative to tree[["where"]]that orders groups from left to right in draw order.Usagegroup.tree(tree)Argumentstree object of class rpart or tree.DetailsNormally used with map.groups.See example.ValueVector of rearranged tree[["where"]]Author(s)Denis WhiteSee Alsorpart,map.groupsExampleslibrary(rpart)data(oregon.env.vars,oregon.grid)group<-group.tree(clip.rpart(rpart(oregon.env.vars),best=7))names(group)<s(oregon.env.vars)map.groups(oregon.grid,group=group)8kgs kgs KGS Measure for Pruning Hierarchical ClustersDescriptionComputes the Kelley-Gardner-Sutcliffe penalty function for a hierarchical cluster tree.Usagekgs(cluster,diss,alpha=1,maxclust=NULL)Argumentscluster object of class hclust or twins.diss object of class dissimilarity or dist.alpha weight for number of clusters.maxclust maximum number of clusters for which to compute measure.DetailsKelley et al.(see reference)proposed a method that can help decide where to prune a hierarchical cluster tree.At any level of the tree the mean across all clusters of the mean within clusters of the dissimilarity measure is calculated.After normalizing,the number of clusters times alpha is added.The minimum of this function corresponds to the suggested pruning size.The current implementation has complexity O(n*n*maxclust),thus very slow with large n.For improvements,at least it should only calculate the spread for clusters that are split at each level, rather than over again for all.ValueVector of the penalty function for trees of size2:maxclust.The names of vector elements are the respective numbers of clusters.Author(s)Denis WhiteReferencesKelley,L.A.,Gardner,S.P.,Sutcliffe,M.J.(1996)An automated approach for clustering an ensem-ble of NMR-derived protein structures into conformationally-related subfamilies,Protein Engineer-ing,9,1063-1065.See Alsotwins.object,dissimilarity.object,hclust,dist,clip.clust,map.groups9Exampleslibrary(cluster)data(votes.repub)a<-agnes(votes.repub,method="ward")b<-kgs(a,a$diss,maxclust=20)plot(names(b),b,xlab="#clusters",ylab="penalty")map.groups Map Groups of ObservationsDescriptionDraws maps of groups of observations created by clustering,classiﬁcation or regression trees,or some other type of classiﬁcation.Usagemap.groups(pts,group,pch=par("pch"),size=2,col=NULL,border=NULL,new=TRUE)Argumentspts matrix or data frame with components"x",and"y"for each observation(see details).group vector of integer class numbers corresponding to pts(see details),and indexing colors in col.pch symbol number from par("pch")if<100,otherwise parameter n for ngon.size size in cex units of point symbol.col vector ofﬁll colors from hsv,rgb,etc,or if NULL,then use rainbow.border vector of border colors from hsv,rgb,etc,or if NULL,then use rainbow.new if TRUE,call plot.new.DetailsIf the number of rows of pts is not equal to the length of group,then(1)pts are assumed to represent polygons and polygon is used,(2)the identiﬁers in group are matched to the polygons in pts through names(group)and pts$x[is.na(pts$y)],and(3)these identiﬁers are mapped to dense integers to reference colours.Otherwise,group is assumed to parallel pts,and,if pch<100, then points is used,otherwise ngon,to draw shaded polygon symbols for each observation in pts. ValueThe vector ofﬁll colors supplied or generated.10map.key Author(s)Denis WhiteSee Alsongon,polygon,group.clust,group.tree,map.keyExamplesdata(s,oregon.env.vars,oregon.bird.dist)data(oregon.border,oregon.grid)#range map for American Avocetspp<-match("American avocet",s[[""]])group<-oregon.bird.dist[,spp]+1names(group)<s(oregon.bird.dist)kol<-gray(seq(0.8,0.2,length.out=length(table(group))))map.groups(oregon.grid,group=group,col=kol)lines(oregon.border)#distribution of January temperaturescuts<-quantile(oregon.env.vars[["jan.temp"]],probs=seq(0,1,1/5))group<-cut(oregon.env.vars[["jan.temp"]],cuts,labels=FALSE,include.lowest=TRUE)names(group)<s(oregon.env.vars)kol<-gray(seq(0.8,0.2,length.out=length(table(group))))map.groups(oregon.grid,group=group,col=kol)lines(oregon.border)#January temperatures using point symbols rather than polygonsmap.groups(oregon.env.vars,group,col=kol,pch=19)lines(oregon.border)map.key Draw Key to accompany Map of GroupsDescriptionDraws legends for maps of groups of observations.Usagemap.key(x,y,labels=NULL,cex=par("cex"),pch=par("pch"),size=2.5*cex,col=NULL,head="",sep=0.25*cex,new=FALSE)map.key11Argumentsx,y coordinates of lower left position of key in proportional units(0-1)of plot.labels vector of labels for classes,or if NULL,then integers1:length(col),or1.size size in cex units of shaded key symbol.pch symbol number for par if<100,otherwise parameter n for ngon.cex pointsize of text,par parameter.head text heading for key.sep separation in cex units between adjacent symbols in key.If sep=0,assume a continuous scale,use square symbols,and put labels at breaks between squares.col vector of colors from hsv,rgb,etc,or if NULL,then use rainbow.new if TRUE,call plot.DetailsUses points or ngon,depending on value of pch,to draw shaded polygon symbols for key. ValueThe vector of colors supplied or generated.Author(s)Denis WhiteSee Alsongon,map.groupsExamplesdata(oregon.env.vars)#key for examples in help(map.groups)#range map for American Avocetkol<-gray(seq(0.8,0.2,length.out=2))map.key(0.2,0.2,labels=c("absent","present"),pch=106,col=kol,head="key",new=TRUE)#distribution of January temperaturescuts<-quantile(oregon.env.vars[["jan.temp"]],probs=seq(0,1,1/5))kol<-gray(seq(0.8,0.2,length.out=5))map.key(0.2,0.2,labels=as.character(round(cuts,0)),col=kol,sep=0,head="key",new=TRUE)#key for example in help file for group.treemap.key(0.2,0.2,labels=as.character(seq(6)),pch=19,head="node",new=TRUE)12ngon ngon Outline or Fill a Regular PolygonDescriptionDraws a regular polygon at speciﬁed coordinates as an outline or shaded.Usagengon(xydc,n=4,angle=0,type=1)Argumentsxydc four element vector with x and y coordinates of center,d diameter in mm,and c color.n number of sides for polygon(>8=>circle).angle rotation angle ofﬁgure,in degrees.type type=1=>interiorﬁlled,type=2=>edge,type=3=>both.DetailsUses polygon to draw shaded polygons and lines for outline.If n is odd,there is a vertex at(0, d/2),otherwise the midpoint of a side is at(0,d/2).ValueInvisible.Author(s)Denis WhiteSee Alsopolygon,lines,map.key,map.groupsExamplesplot(c(0,1),c(0,1),type="n")ngon(c(.5,.5,10,"blue"),angle=30,n=3)apply(cbind(runif(8),runif(8),6,2),1,ngon)oregon.bird.dist13 oregon.bird.dist Presence/Absence of Bird Species in Oregon,USADescriptionBinary matrix(1=present)for distributions of248native breeding bird species for389grid cells in Oregon,USA.Usagedata(oregon.bird.dist)FormatA data frame with389rows and248columns.DetailsRow names are hexagon identiﬁers from White et al.(1992).Column names are species ele-ment codes developed by The Nature Conservancy(TNC),the Oregon Natural Heritage Program (ONHP),and NatureServe.SourceDenis WhiteReferencesMaster,L.(1996)Predicting distributions for vertebrate species:some observations,Gap Analysis:A Landscape Approach to Biodiversity Planning,Scott,J.M.,Tear,T.H.,and Davis,F.W.,editors,American Society for Photogrammetry and Remote Sensing,Bethesda,MD,pp.171-176.White,D.,Preston,E.M.,Freemark,K.E.,Kiester,A.R.(1999)A hierarchical framework for con-serving biodiversity,Landscape ecological analysis:issues and applications,Klopatek,J.M.,Gard-ner,R.H.,editors,Springer-Verlag,pp.127-153.White,D.,Kimerling,A.J.,Overton,W.S.(1992)Cartographic and geometric components of a global sampling design for environmental monitoring,Cartography and Geographic Information Systems,19(1),5-22.TNC,https:///en-us/ONHP,https:///orbic/NatureServe,https:///See Alsooregon.env.vars,s,oregon.grid,oregon.borders s Names of Bird Species in Oregon,USADescriptionScientiﬁc and common names for248native breeding bird species in Oregon,USA.Usagedata(s)FormatA data frame with248rows and2columns.DetailsRow names are species element codes.Columns are""and"".Data are provided by The Nature Conservancy(TNC),the Oregon Natural Heritage Program(ONHP), and NatureServe.SourceDenis WhiteReferencesMaster,L.(1996)Predicting distributions for vertebrate species:some observations,Gap Analysis:A Landscape Approach to Biodiversity Planning,Scott,J.M.,Tear,T.H.,and Davis,F.W.,editors,American Society for Photogrammetry and Remote Sensing,Bethesda,MD,pp.171-176.TNC,https:///en-us/ONHP,https:///orbic/NatureServe,https:///See Alsooregon.bird.distoregon.border15 oregon.border Boundary of Oregon,USADescriptionThe boundary of the state of Oregon,USA,in lines format.Usagedata(oregon.border)FormatA data frame with485rows and2columns(the components"x"and"y").DetailsThe map projection for this boundary,as well as the point coordinates in oregon.env.vars,is the Lambert Conformal Conic with standard parallels at33and45degrees North latitude,with the longitude of the central meridian at120degrees,30minutes West longitude,and with the projection origin latitude at41degrees,45minutes North latitude.SourceDenis Whiteoregon.env.vars Environmental Variables for Oregon,USADescriptionDistributions of10environmental variables for389grid cells in Oregon,USA.Usagedata(oregon.env.vars)FormatA data frame with389rows and10columns.16oregon.gridDetailsRow names are hexagon identiﬁers from White et al.(1992).Variables(columns)arebird.spp number of native breeding bird speciesx x coordinate of center of grid celly y coordinate of center of grid celljan.temp mean minimum January temperature(C)jul.temp mean maximum July temperature(C)rng.temp mean difference between July and January temperatures(C)ann.ppt mean annual precipitation(mm)min.elev minimum elevation(m)rng.elev range of elevation(m)max.slope maximum slope(percent)SourceDenis WhiteReferencesWhite,D.,Preston,E.M.,Freemark,K.E.,Kiester,A.R.(1999)A hierarchical framework for con-serving biodiversity,Landscape ecological analysis:issues and applications,Klopatek,J.M.,Gard-ner,R.H.,editors,Springer-Verlag,pp.127-153.White,D.,Kimerling,A.J.,Overton,W.S.(1992)Cartographic and geometric components of a global sampling design for environmental monitoring,Cartography and Geographic Information Systems,19(1),5-22.See Alsooregon.bird.dist,oregon.grid,oregon.borderoregon.grid Hexagonal Grid Cell Polygons covering Oregon,USADescriptionPolygon borders for389hexagonal grid cells covering Oregon,USA,in polygon format.Usagedata(oregon.grid)FormatA data frame with3112rows and2columns(the components"x"and"y").DetailsThe polygon format used for these grid cell boundaries is a slight variation from the standard R/S format.Each cell polygon is described by seven coordinate pairs,the last repeating theﬁrst.Prior to theﬁrst coordinate pair of each cell is a row containing NA in the"y"column and,in the"x"col-umn,an identiﬁer for the cell.The identiﬁers are the same as the row names in oregon.bird.dist and oregon.env.vars.See map.groups for how the linkage is made in mapping.These grid cells are extracted from a larger set covering the conterminous United States and adjacent parts of Canada and Mexico,as described in White et al.(1992).Only cells with at least50percent of their area contained within the state of Oregon are included.The map projection for the coordinates,as well as the point coordinates in oregon.env.vars,is the Lambert Conformal Conic with standard parallels at33and45degrees North latitude,with the longitude of the central meridian at120degrees,30minutes West longitude,and with the projection origin latitude at41degrees,45minutes North latitude.SourceDenis WhiteReferencesWhite,D.,Kimerling,A.J.,Overton,W.S.(1992)Cartographic and geometric components of a global sampling design for environmental monitoring,Cartography and Geographic Information Systems,19(1),5-22.twins.to.hclust Converts agnes or diana object to hclust objectDescriptionAlternative to as.hclust that retains cluster data.Usagetwins.to.hclust(cluster)Argumentscluster object of class twins.DetailsUsed internally in with clip.clust and draw.clust.Valuehclust objectAuthor(s)Denis WhiteSee Alsohclust,twins.objectIndex∗aplotmap.key,10ngon,12∗clusterclip.clust,2clip.rpart,3draw.clust,4group.clust,6kgs,8map.groups,9twins.to.hclust,17∗datasetsoregon.bird.dist,13s,14oregon.border,15oregon.env.vars,15oregon.grid,16∗hplotdraw.clust,4draw.tree,5map.groups,9map.key,10∗manipclip.clust,2clip.rpart,3group.clust,6group.tree,7kgs,8twins.to.hclust,17∗treedraw.tree,5group.tree,7map.groups,9 agnes,4as.hclust,17clip.clust,2,8,17clip.rpart,3cutree,2,6,7diana,4dissimilarity.object,8dist,8draw.clust,2,4,6,17draw.tree,4,5group.clust,6,10group.tree,7,10hclust,2,4,7,8,18hsv,4,5,9,11kgs,8lines,12,15map.groups,4,6,7,9,11,12,17map.key,10,10,12ngon,9–11,12oregon.bird.dist,13,14,16,17s,13,14oregon.border,13,15,16oregon.env.vars,13,15,15,17oregon.grid,13,16,16par,11plot,11plot.hclust,4plot.new,4,5,9pltree,4points,9,11polygon,9,10,12,16prune.rpart,3rainbow,4,5,9,11rgb,4,5,9,11rpart,3,6,7twins.object,2,7,8,18twins.to.hclust,1719。

基于分形理论的桂中盆地典型成矿元素地球化学异常提取

基于分形理论的桂中盆地典型成矿元素地球化学异常提取刘舒飞; 王佐满【期刊名称】《《中国矿业》》【年(卷),期】2019(028)0z2【总页数】5页(P258-262)【关键词】桂中盆地; 异常提取; 分形; 多重分形【作者】刘舒飞; 王佐满【作者单位】中国黄金集团有限公司北京100011【正文语种】中文【中图分类】P632桂中盆地位于扬子板块和华夏板块相邻位置，区内成矿以锰、锡及铜矿床为主，且矿床类型多样，其中锰矿床主要为风化堆积锰矿床和沉积锰矿床，锡矿床则以丹池一带的锡多金属矿床和靠近大瑶山附近的锡钨及锡钽铌共生矿床为主。

根据1∶20万水系沉积物化探数据，盆地内Mn、Sn元素的异常连续性较好，且存在明显的浓度梯度变化。

但受区内矿床类型较多并分布广泛的影响，元素富集存在一定的奇异性和复杂性，部分区域成矿元素化探异常与矿床分布嵌合不明显，传统化探方法的异常提取存在一定的局限。

与成矿元素有关的区域地球化学数据普遍不符合正态分布，但具备分形和多重分形的特征，近年来不少学者利用分形的方法来研究元素分布、确定异常下限并提取各类矿致异常或者弱隐异常、重叠异常，取得较好效果。

常用的分形方法有含量-面积模型(C-A)、能谱-面积模型(S-A)、多重分形谱及元素奇异性指数等[1-8]。

笔者利用桂中盆地Mn、Sn元素地球化学数据，结合分形理论，研究了锰、锡元素分布特征和富集规律，并探讨了不同类型矿床的成矿特征。

1 区域地质背景桂中盆地西缘以南丹-昆仑关断裂带与右江盆地相邻，北侧以宜山-柳城弧形断裂带与扬子地台江南造山带西段隔开，东侧过渡为大瑶山隆起区。

盆地内主要出露地层为泥盆系和石炭系，主要以泥盆系为主，石炭系仅在盆地中部和周缘出露。

主要发育泥质灰岩、生物灰岩、白云质灰岩等碳酸岩及含泥质砂岩、粉砂岩等。

盆地内构造主要以平缓开阔的褶皱为主，断裂发育较少，仅在局部有小型逆掩断层。

桂中盆地周缘岩浆岩分布较多，如昆仑关岩体、龙头山岩体、花山-姑婆山岩体等，这一系列岩体与盆地周缘的钨、锡、金、铜等矿床的产出有密切关系。