Lecture Notes in Computer Science 1 Grid Computing on the Web Using the Globus Toolkit
Springer Lecture Notes 论文格式
Lecture Notes in Computer Science:Authors’ Instructions for the Preparationof Camera-Ready Contributionsto LNCS/LNAI/LNBI ProceedingsAlfred Hofmann1,1, Brigitte Apfel1, Ursula Barth1, Christine Günther1, Ingrid Haas1, Frank Holzwarth1, Anna Kramer1, Leonie Kunz1,Nicole Sator1, Erika Siebert-Cole1 and Peter Straßer1,1 Springer-Verlag, Computer Science Editorial, Tiergartenstr. 17,69121 Heidelberg, Germany{Alfred.Hofmann, Brigitte.Apfel, Ursula.Barth, Christine.Guenther,Ingrid.Haas, Frank.Holzwarth, Anna.Kramer, Leonie.Kunz,Nicole.Sator, Erika.Siebert-Cole, Peter.Strasser, LNCS}@ Abstract. The abstract should summarize the contents of the paper and shouldcontain at least 70 and at most 150 words. It should be set in 9-point font sizeand should be inset 1.0 cm from the right and left margins. There should be twoblank (10-point) lines before and after the abstract. This document is in therequired format.Keywords: We would like to encourage you to list your keywords in thissection.1 IntroductionThis instruction file for Word users (there is a separate instruction file for LaTeX users) may be used as a template. Kindly send the final and checked Word and PDF files of your paper to the Contact Volume Editor. This is usually one of the organizers of the conference. You should make sure that the Word and the PDF files are identical and correct and that only one version of your paper is sent. It is not possible to update files at a later stage. Please note that we do not need the printed paper.We would like to draw your attention to the fact that it is not possible to modify a paper in any way, once it has been published. This applies to both the printed book and the online version of the publication. Every detail, including the order of the names of the authors, should be checked before the paper is sent to the Volume Editors.1Please note that the LNCS Editorial assumes that all authors have used the westernnaming convention, with given names preceding surnames. This determines thestructure of the names in the running heads and the author index.1.1 Checking the PDF FileKindly assure that the Contact Volume Editor is given the name and email address of the contact author for your paper. The Contact Volume Editor uses these details to compile a list for our production department at SPS in India. Once the files have been worked upon, SPS sends a copy of the final pdf of each paper to its contact author. The contact author is asked to check through the final pdf to make sure that no errors have crept in during the transfer or preparation of the files. This should not be seen as an opportunity to update or copyedit the papers, which is not possible due to time constraints. Only errors introduced during the preparation of the files will be corrected.This round of checking takes place about two weeks after the files have been sent to the Editorial by the Contact Volume Editor, i.e., roughly seven weeks before the start of the conference for conference proceedings, or seven weeks before the volume leaves the printer’s, for post-proceedings. If SPS does not receive a reply from a particular contact author, within the timeframe given, then it is presumed that the author has found no errors in the paper. The tight publication schedule of LNCS does not allow SPS to send reminders or search for alternative email addresses on the Internet.In some cases, it is the Contact Volume Editor that checks all the pdfs. In such cases, the authors are not involved in the checking phase.1.2 Additional Information Required by the Volume EditorIf you have more than one surname, please make sure that the Volume Editor knows how you are to be listed in the author index.1.3 Copyright FormsThe copyright form may be downloaded from the For Authors section of the LNCS Webpage: /lncs. Please send your signed copyright form to the Contact Volume Editor, either as a scanned pdf or by fax or by courier. One author may sign on behalf of all of the other authors of a particular paper. Digital signatures are acceptable.2 Paper PreparationThe printing area is 122 mm × 193 mm. The text should be justified to occupy the full line width, so that the right margin is not ragged, with words hyphenated as appropriate. Please fill pages so that the length of the text is no less than 180 mm, if possible.Use 10-point type for the name(s) of the author(s) and 9-point type for the address(es) and the abstract. For the main text, please use 10-point type and single-line spacing. We recommend the use of Computer Modern Roman or Times. Italictype may be used to emphasize words in running text. Bold type and underlining should be avoided.Papers not complying with the LNCS style will be reformatted. This can lead to an increase in the overall number of pages. We would therefore urge you not to squash your paper.Headings.Headings should be capitalized (i.e., nouns, verbs, and all other words except articles, prepositions, and conjunctions should be set with an initial capital) and should, with the exception of the title, be aligned to the left. Words joined by a hyphen are subject to a special rule. If the first word can stand alone, the second word should be capitalized. The font sizes are given in Table 1.Here are some examples of headings: "Criteria to Disprove Context-Freeness of Collage Languages", "On Correcting the Intrusion of Tracing Non-deterministic Programs by Software", "A User-Friendly and Extendable Data Distribution System", "Multi-flip Networks: Parallelizing GenSAT", "Self-determinations of Man".Table 1.Font sizes of headings. Table captions should always be positioned above the tables.Heading level Example Font size and styleTitle (centered) Lecture Notes …14 point, bold1st-level heading 1 Introduction12 point, bold2nd-level heading 2.1 Printing Area10 point, bold3rd-level heading Headings.Text fol lows …10 point, bold4th-level heading Remark.Text follows …10 point, italic Lemmas, Propositions, and Theorems.The numbers accorded to lemmas, propositions, and theorems, etc. should appear in consecutive order, starting with Lemma 1, and not, for example, with Lemma 11.2.1 FiguresPlease check that the lines in line drawings are not interrupted and have a constant width. Grids and details within the figures must be clearly legible and may not be written one on top of the other. Line drawings should have a resolution of at least 800 dpi (preferably 1200 dpi). The lettering in figures should have a height of 2 mm (10-point type). Figures should be numbered and should have a caption which should always be positioned under the figures, in contrast to the caption belonging to a table, which should always appear above the table. Please center the captions between the margins and set them in 9-point type (Fig. 1 shows an example). The distance between text and figure should be about 8 mm, the distance between figure and caption about 6 mm.To ensure that the reproduction of your illustrations is of a reasonable quality, we advise against the use of shading. The contrast should be as pronounced as possible.If screenshots are necessary, please make sure that you are happy with the print quality before you send the files.Remark 1. In the printed volumes, illustrations are generally black and white (halftones), and only in exceptional cases, and if the author is prepared to cover the extra costs involved, are colored pictures accepted. Colored pictures are welcome in the electronic version free of charge. If you send colored figures that are to be printed in black and white, please make sure that they really are legible in black and white. Some colors show up very poorly when printed in black and white.Fig. 1. One kernel at x s (dotted kernel) or two kernels at x i and x j (left and right) lead to the same summed estimate at x s. This shows a figure consisting of different types of lines. Elements of the figure described in the caption should be set in italics, in parentheses, as shown in this sample caption.2.2 FormulasDisplayed equations or formulas are centered and set on a separate line (with an extra line or halfline space above and below). Displayed expressions should be numbered for reference. The numbers should be consecutive within each section or within the contribution, with numbers enclosed in parentheses and set on the right margin.x + y = z . (1) Equations should be punctuated in the same way as ordinary text but with a small space before the end punctuation mark.2.3 FootnotesThe superscript numeral used to refer to a footnote appears in the text either directly after the word to be discussed or – in relation to a phrase or a sentence – following the punctuation mark (comma, semicolon, or period). Footnotes should appear at the bottom of the normal text area, with a line of about 5cm set immediately above them2.2.4 Program CodeProgram listings or program commands in the text are normally set in typewriter font, e.g., CMTT10 or Courier.Example of a Computer Program from Jensen K., Wirth N. (1991) Pascal user manual and report. Springer, New Yorkprogram Inflation (Output){Assuming annual inflation rates of 7%, 8%, and10%,... years};const MaxYears = 10;var Year: 0..MaxYears;Factor1, Factor2, Factor3: Real;beginYear := 0;2The footnote numeral is set flush left and the text follows with the usual word spacing.Factor1 := 1.0; Factor2 := 1.0; Factor3 := 1.0;WriteLn('Year 7% 8% 10%'); WriteLn;repeatYear := Year + 1;Factor1 := Factor1 * 1.07;Factor2 := Factor2 * 1.08;Factor3 := Factor3 * 1.10;WriteLn(Year:5,Factor1:7:3,Factor2:7:3,Factor3:7:3)until Year = MaxYearsend.2.5 CitationsFor citations in the text please use square brackets and consecutive numbers: [1], [2], [3], etc.2.6 Page Numbering and Running HeadsThere is no need to include page numbers. If your paper title is too long to serve as a running head, it will be shortened. Your suggestion as to how to shorten it would be most welcome.3 LNCS OnlineThe online version of the volume will be available in LNCS Online. Members of institutes subscribing to the Lecture Notes in Computer Science series have access to all the pdfs of all the online publications. Non-subscribers can only read as far as the abstracts. If they try to go beyond this point, they are automatically asked, whether they would like to order the pdf, and are given instructions as to how to do so.Please note that, if your email address is given in your paper, it will also be included in the meta data of the online version.4 BibTeX EntriesThe correct BibTeX entries for the Lecture Notes in Computer Science volumes can be found at the following Website shortly after the publication of the book: rmatik.uni-trier.de/~ley/db/journals/lncs.html Acknowledgments. The heading should be treated as a 3rd level heading and should not be assigned a number.5 The References SectionIn order to permit cross referencing within LNCS-Online, and eventually between different publishers and their online databases, LNCS will, from now on, be standardizing the format of the references. This new feature will increase the visibility of publications and facilitate academic research considerably. Please base your references on the examples below. References that don’t adhere to this style will be reformatted by Springer. You should therefore check your references thoroughly when you receive the final pdf of your paper. The reference section must be complete. You may not omit references. Instructions as to where to find a fuller version of the references are not permissible.We only accept references written using the latin alphabet. If the title of the book you are referring to is in Russian or Chinese, then please write (in Russian) or (in Chinese) at the end of the transcript or translation of the title.The following section shows a sample reference list with entries for journal articles [1], an LNCS chapter [2], a book [3], proceedings without editors [4] and [5], as well as a URL [6]. Please note that proceedings published in LNCS are not cited with their full titles, but with their acronyms!References1. Smith, T.F., Waterman, M.S.: Identification of Common Molecular Subsequences. J. Mol.Biol. 147, 195--197 (1981)2. May, P., Ehrlich, H.C., Steinke, T.: ZIB Structure Prediction Pipeline: Composing aComplex Biological Workflow through Web Services. In: Nagel, W.E., Walter, W.V., Lehner, W. (eds.) Euro-Par 2006. LNCS, vol. 4128, pp. 1148--1158. Springer, Heidelberg (2006)3. Foster, I., Kesselman, C.: The Grid: Blueprint for a New Computing Infrastructure. MorganKaufmann, San Francisco (1999)4. Czajkowski, K., Fitzgerald, S., Foster, I., Kesselman, C.: Grid Information Services forDistributed Resource Sharing. In: 10th IEEE International Symposium on High Performance Distributed Computing, pp. 181--184. IEEE Press, New York (2001)5. Foster, I., Kesselman, C., Nick, J., Tuecke, S.: The Physiology of the Grid: an Open GridServices Architecture for Distributed Systems Integration. Technical report, Global Grid Forum (2002)6. National Center for Biotechnology Information, Appendix: Springer-Author DiscountLNCS authors are entitled to a 33.3% discount off all Springer publications. Before placing an order, they should send an email to **********************************, giving full details of their Springer publication, to obtain a so-called token. This token is a number, which must be entered when placing an order via the Internet, in order to obtain the discount.Checklist of Items to be Sent to Volume Editors1.A final Word or RTF file2.A final PDF file3.A copyright form, signed by one author on behalf of all of the authors of the paper4.A readme giving the name and email address of the corresponding author。
整理电子课本下载_电子书类型
整理人 尼克电子课本下载附件1江西省中小学电子书包应用研究方案一、课题提出1.江西省中小学教育信息化背景。
近年来，全省各地“班班通”建设不断推广，南昌、新余等设区市大部分学校实现班班通；农村义务教育薄弱学校改造项目、教学点数字教育资源全覆盖项目如火似荼。
江西省基础教育资源网发展至今拥有中小学各年级各类型教育资源总量达167万余条，资源容量有50T。
省电教馆每年举办中小学教师“班班通”教学资源应用大赛等9项应用展示活动，“十二五”期间全省中小学、幼儿园教育技术研究立项课题达662个，营造并形成了信息化教学的良好氛围。
“智慧校园”、“省级教育资源公共服务平台”的兴起与发展，为电子书包的应用提供了充分的技术与环境支撑。
从教育云的视角看，“三通两平台”的建设与应用，具体体现在中小学校学生的学习活动之中，就是移动学习终端的引入和使用。
电子书包作为中小学学生的移动学习终端，实质上是“人人通”的直接入口，是一种个人学习的信息化环境，其教育教学应用自然是顺应数字时代发展，数字化校园建设和教学方式改革的必然选择。
2.国内外同类研究综述从中国万方数据的知识脉络检索（2014.03.09.），获得电子书包的研究趋势（图1）：图1 电子书包的研究趋势（1）国际教育信息化发展趋势。
21世纪初，八国集团在冲绳发表的《全球信“信息通信技术是21世纪社会发展的最强有力动力之一，息社会冲绳宪章》中认为：并将迅速成为世界经济增长的重要动力。
”美国于2010年11月颁布最新一轮的《国家教育技术计划》中，通过对现有的信息化教育模式进行反思后提出“以技术支持的教育系统结构性变革”。
当前，以美国、韩国、新加坡、日本为首的世界各国“一对一”数字化教育应用项目正在迅猛发展。
在这些“一对一”数字化项目中，电子书包项目表现得尤为突出。
据克利夫兰市场咨询公司的调查报告迄今至少有50个国家（地区）计划推广电子课本、电子书包。
电子书包在世界很多国家都被提上了政府议程，同时也得到了企业和社会的纷纷关注。
Springer论文格式
Author Guidelines for the Preparation of Contributions to Springer Computer Science Proceedings Alfred Hofmann1,*, Ralf Gerstner1, Anna Kramer1, and Frank Holzwarth21 Springer-Verlag, Computer Science Editorial, Heidelberg, Germany{alfred.hofmann,ralf.gerstner,anna.kramer}@2 Springer-Verlag, Technical Support, Heidelberg, Germanyfrank.holzwarth@Abstract. The abstract is a mandatory element that should summarize the con-tents of the paper and should contain at least 70 and at most 150 words. Ab-stract and keywords are freely available in SpringerLink.Keywords: We would like to encourage you to list your keywords here. Theyshould be separated by middots.1IntroductionYou will find here Springer’s guidelines for the preparation of proceedings papers to be published in one of the following series, in printed and electronic form:∙Lecture Notes in Computer Science (LNCS), incl. its subseries Lecture Notes in Artificial Intelligence (LNAI) and Lecture Notes in Bioinformatics (LNBI), and LNCS Transactions;∙Lecture Notes in Business Information Processing (LNBIP);∙Communications in Computer and Information Science (CCIS);∙Lecture Notes of the Institute for Computer Sciences, Social Informatics and Tele-communications Engineering (LNICST);∙IFIP Advances in Information and Communication Technology (IFIP AICT), for-merly known as the IFIP Series;∙Proceedings in Information and Communication Technology (PICT).Your contribution may be prepared in LaTeX or Microsoft Word. Technical Instruc-tions for working with Springer’s style files and templates are provided in separate documents which can be found in the respective zip packages on our website.*No academic titles or descriptions of academic positions should be included in the addresses.The affiliations should consist of the author’s institution, town, and country.22Preparation of Your Paper2.1Structuring Your PaperAffiliations. The affiliated institutions are to be listed directly below the names of the authors. Multiple affiliations should be marked with superscript Arabic numbers, and they should each start on a new line as shown in this document. In addition to the name of your affiliation, we would ask you to give the town and the country in which it is situated. If you prefer to include the entire postal address, then please feel free to do so. E-mail addresses should start on a new line and should be grouped per affiliation. Headings.Headings should be capitalized (i.e., nouns, verbs, and all other words except articles, prepositions, and conjunctions should be set with an initial capital) and should, with the exception of the title, be aligned to the left. Only the first two levels of section headings should be numbered, as shown in Table 1. The respective font sizes are also given in Table 1. Kindly refrain from using “0” when numbering your section headings.Table 1. Font sizes of headings. Table captions should always be positioned above the tables. Heading level Example Font size and style Title (centered) Lecture Notes14 point, bold1st-level heading 1 Introduction12 point, bold2nd-level heading 2.1 Printing Area10 point, bold3rd-level heading Run-in Heading in Bold. Text follows10 point, bold4th-level heading Lowest Level Heading. Text follows10 point, italic Words joined by a hyphen are subject to a special rule. If the first word can stand alone, the second word should be capitalized.Here are some examples of headings: “Criteria to Disprove Context-Freeness of Collage Languages”, “On Correcting the Intrusion of Tracing Non-deterministic Pro-grams by Software”, “A User-Friendly and Extendable Data Distribution System”, “Multi-flip Networks: Parallelizing GenSAT”, “Self-determinations of Man”. Lemmas, Propositions, and Theorems.The numbers accorded to lemmas, proposi-tions, and theorems, etc. should appear in consecutive order, starting with Lemma 1. Please do not include section counters in the numbering like “Theorem 1.1”.2.2Length of PapersWe only wish to publish papers of significant scientific content. Very short papers will be moved to the back matter, will not be made available for indexing, and will not be visible as individual papers on SpringerLink.32.3 Page Numbering and Running HeadsThere is no need to include page numbers or running heads; this will be done at ourend. If your paper title is too long to serve as a running head, it will be shortened.Your suggestion as to how to shorten it would be most welcome.2.4 Figures and TablesIt is essential that all illustrations are clear and legible. Vector graphics (rather thanrasterized images) should be used for diagrams and schemas whenever possible.Please check that the lines in line drawings are not interrupted and have a constantwidth. Grids and details within the figures must be clearly legible and may not bewritten one on top of the other. Line drawings are to have a resolution of at least 800dpi (preferably 1200 dpi). The lettering in figures should not use font sizesp o w e r f l u c t u a t i o n (1555 n m ), d B time, msFig. 1. Power distribution of channel at 1555 nm along the link of 383 km (Source: LNCS5412, p. 323)Fig. 2. Artifacts empowered by Artificial Intelligence (Source: LNCS 5640, p. 115)4smaller than 6 pt (~ 2mm character height). Figures are to be numbered and to have acaption which should always be positioned under the figures, in contrast to the captionbelonging to a table, which should always appear above the table.Captions are set in 9-point type. If they are short, they are centered between themargins. Longer captions, covering more than one line, are justified (Fig. 1 and Fig. 2show examples). Captions that do not constitute a full sentence, do not have a period.Text fragments of fewer than four lines should not appear at the tops or bottoms ofpages, following a table or figure. In such cases, it is better to set the figures right atthe top or right at the bottom of the page.If screenshots are necessary, please make sure that the essential content is clear tothe reader.Remark 1. In the printed volumes, illustrations are generally black and white (half-tones), and only in exceptional cases, and if the author or the conference organizationis prepared to cover the extra costs involved, are colored pictures accepted. Coloredpictures are welcome in the electronic version free of charge. If you send coloredfigures that are to be printed in black and white, please make sure that they really arealso legible in black and white. Some colors show up very poorly when printed inblack and white.2.5FormulasDisplayed equations or formulas are centered and set on a separate line (with an extraline or half line space above and below). Displayed expressions should be numberedfor reference. The numbers should be consecutive within the contribution, with num-bers enclosed in parentheses and set on the right margin. Please do not include sectioncounters in the numbering.x + y = z (1) Equations should be punctuated in the same way as ordinary text but with a smallspace before the end punctuation mark.2.6FootnotesThe superscript numeral used to refer to a footnote appears in the text either directlyafter the word to be discussed or – in relation to a phrase or a sentence – following thepunctuation mark (comma, semicolon, or period).1For remarks pertaining to the title or the authors’ names, in the header of a paper,symbols should be used instead of a number (see first page of this document). Pleasenote that no footnotes may be included in the abstract.1The footnote numeral is set flush left and the text follows with the usual word spacing.5 2.7Program CodeProgram listings or program commands in the text are normally set in typewriter font: program Inflation (Output){Assuming annual inflation rates of 7%, 8%, and10%,... years};const MaxYears = 10;var Year: 0..MaxYears;Factor1, Factor2, Factor3: Real;beginYear := 0;Factor1 := 1.0; Factor2 := 1.0; Factor3 := 1.0;WriteLn('Year 7% 8% 10%'); WriteLn;repeatYear := Year + 1;Factor1 := Factor1 * 1.07;Factor2 := Factor2 * 1.08;Factor3 := Factor3 * 1.10;WriteLn(Year:5,Factor1:7:3,Factor2:7:3,Factor3:7:3)until Year = MaxYearsend.[Example of a computer program from Jensen K., Wirth N.: Pascal User Manual and Report. Springer, New York (1991)]2.8Citations and BibliographyFor citations in the text, please use square brackets and consecutive numbers. We would write [1,2,3,4,5] for consecutive numbers and [1], [3], [5] for non-consecutive numbers. The numbers in the bibliography section are without square brackets. We prefer numbered references to other styles of references, such as those with abbreviat-ed names and years.Please write all references using the Latin alphabet. If the title of the book you are referring to is, e.g., in Russian or Chinese, then please write (in Russian) or (in Chi-nese) at the end of the transcript or translation of the title.In order to permit cross referencing within SpringerLink, and eventually between different publishers and their online databases, Springer standardizes the format of the references. This feature aims to increase the visibility of publications and facilitate academic research. Please base your references on the examples given in the refer-ences section of these instructions. References that do not adhere to this style will be reformatted at our end.We would like to draw your attention to the fact that references to LNCS proceed-ings papers are particularly often reformatted due to missing editor names or incom-plete publisher information. This adjustment may result in the final papers as pub-6lished by Springer having more pages than the original versions as submitted by the authors. Here is an example:Reference as formatted in author’s original version:Assemlal, H.E., Tschumperlé, D., Brun, L.: Efficient Computation of PDF-Based Characteris-tics from Diffusion MR Signal. In: MICCAI. Volume 5242. (2008) 70–78Reference after reformatting by Springer:Assemlal, H.E., Tschumperlé, D., Brun, L.: Efficient Computation of PDF-Based Characteris-tics from Diffusion MR Signal. In: Metaxas, D., Axel, L., Fichtinger, G., Székely, G. (eds.) MICCAI 2008, Part II. LNCS, vol. 5242, pp. 70–78. Springer, Heidelberg (2008)One more line is needed for this reference, as a result of Springer’s adjustment.Please make sure that all your sources are correctly listed in the reference section. Do not include references to pieces of work that are not connected with your paper.The references section at the end of this paper shows a sample reference list with entries for journal articles [1], an LNCS chapter [2], a book [3], proceedings without editors [4] and [5], as well as a URL [6]. Please note that Springer proceedings are cited with their publication acronyms and volume numbers.2.9PlagiarismSpringer takes plagiarism seriously. If an author has copied from another author or has used parts of another author’s work (text, tables, figures, etc.), without his or her permission and a reference, then the paper on SpringerLink will be given a “retracted” stamp, and an erratum explaining the reasons for the retraction will be included. In addition, the volume editors and the author’s academic supervisors will be informed that plagiarism has been committed. Please note that a retracted paper remains visible, with its “retracted” stamp. It does not simply disappear.Acknowledgements. This should always be a run-in heading and not a section or subsection heading. It should not be assigned a number. The acknowledgements may include reference to grants or supports received in relation to the work presented in the paper.3Additional Information Required from Authors3.1Copyright FormThere are different copyright forms in place for the different Springer Computer Sci-ence proceedings book series. A prefilled copyright form is usually available from the conference website. Please send your signed copyright form to your conference publi-cation contact, either as a scanned PDF or by fax or by courier. One author may sign7on behalf of all of the other authors of a particular paper. In this case, the author signs for and accepts responsibility for releasing this material on behalf of any and all co-authors. Digital signatures are not acceptable.3.2Contact Author InformationKindly assure that, when you submit the final version of your paper, you also provide the name and e-mail address of the contact author for your paper. These details are used to compile a list of contact authors for our typesetting partners SPS in India. The contact author must be available to check the paper roughly seven weeks before the start of the conference, or before the book is due to leave the printing office, in the case of post-conference proceedings. Please also mark the corresponding author in the header of the paper, preferably with a small envelope. If this is not done by you, it will be carried out by our typesetters.3.3Correct Representation of Author NamesAuthors’ names should be written out in full at the tops of the papers. They are short-ened by us to “initials surname” in the running heads and take the form “surname, given name” in the author index. If you or any of your co-authors have more than one family name, it should be made quite clear how your name is to be displayed in the running heads and the author index. Chinese authors should write their given names in front of their surnames at the tops of their papers. If you only have one (main) name, please make sure that this name is written out in full in the running heads, when you check your final PDF. Names and affiliations cannot be changed once a paper has been published.4Typesetting of Your Paper at SpringerPlease make sure that the paper you submit is final and complete, that any copyright issues have been resolved, that the authors listed at the top of the chapter really are the final authors, and that you have not omitted any references. Following publication, it is not possible to alter or withdraw your paper on SpringerLink. Kindly note that we prefer the use of American English.4.1What Will Be Done with Your PaperIf the instructions have been followed closely, then only very minor alterations will be made to your paper. The production team at SPS checks the format of the paper, and if, for example, vertical spacing has been inserted or removed, then this is remedied. In addition, running-heads, final page numbers, and a copyright line are inserted, and the capitalization of the headings is checked and corrected if need be. Finally, the refer-8ence section is attuned to our specifications (see also Section 2.7). Light technical copyediting may also be performed.4.2Proof Reading StageOnce the files have been worked upon, SPS sends a copy of the final PDF of each paper to its contact author. The contact author is asked to check through the final PDF to make sure that no errors have crept in during the transfer or preparation of the files. This should not be seen as an opportunity to update or copyedit the paper, which is not possible due to time constraints. Only errors introduced during the preparation of the files will be corrected. Particular attention should be paid to the references section.If SPS does not receive a reply from a particular contact author, within the timeframe given (usually 72 hours), then it is presumed that the author has found no errors in the paper. The tight publication schedule of our proceedings series does not allow SPS to send reminders or search for alternative e-mail addresses on the Internet.In some cases, it is the contact volume editor or the publication chair who checks all of the PDFs. In such cases, the authors are not involved in the checking phase.The purpose of the proof is to check for typesetting or conversion errors and the completeness and accuracy of the text, tables, and figures. Substantial changes in content, e.g., new results, corrected values, title and authorship, are not possible and cannot be processed.5Online Publication in SpringerLinkAll papers are published in our digital library, SpringerLink. Only subscribers to Springer’s eBook packages or to the electronic book series are able to access the full text PDFs of our online publications. Front and back matter, as well as abstracts and references, are freely available for all users.6Checklist of Items to Be Sent to Volume Editor∙The final source files, incl. any non-standard fonts.∙ A final PDF file corresponding exactly to the final source files.∙ A copyright form, signed by one author on behalf of all of the authors of the paper. ∙The name and e-mail address of the contact author who will check the proof of the paper.∙ A suggestion for an abbreviated running head, if appropriate.∙Information about correct representation of authors’ names, where necessary.9 Contact UsIf you have any further questions regarding the preparation of your paper, then please do not hesitate to get in touch with us.∙For all questions related to our LaTeX style files, your contact person is: Mr. Frank Holzwarth, e-mail: frank.holzwarth@.∙For overall technical questions concerning the preparation of LNCS/LNAI/LNBI papers, please contact Ms. Anna Kramer, e-mail: lncs@.∙For the LNBIP series, please contact Ms. Viktoria Meyer, e-mail: lnbip@.∙For the CCIS series, please contact Ms. Leonie Kunz, e-mail:ccis@.∙For the LNICST series, please contact Mr. Peter Strasser, e-mail: lnicst@.∙For the IFIP AICT series, please contact Ms. Erika Siebert-Cole, e-mail: ifip@.Springer Author DiscountAuthors contributing to any of Springer’s Computer Science proceedings publications are entitled to a 33.3% discount off all Springer products when placing an order through . To make use of this discount, please access the following page: /gp/authors-editors/book-authors-editors/springertoken-request-for-springer-authors/4090. You will be requested to give full details of your Springer publication and will be given a so-called SpringerToken. This token is a number that must be entered when placing an order through , in order to obtain the discount.A AppendixThe appendix should be positioned in front of the references. If it has been placed elsewhere, it will be moved by our typesetters.References1.Smith, T.F., Waterman, M.S.: Identification of Common Molecular Subsequences. J. Mol.Biol. 147, 195–197 (1981)2.May, P., Ehrlich, H.C., Steinke, T.: ZIB Structure Prediction Pipeline: Composing a Com-plex Biological Workflow through Web Services. In: Nagel, W.E., Walter, W.V., Lehner, W. (eds.) Euro-Par 2006. LNCS, vol. 4128, pp. 1148–1158. Springer, Heidelberg (2006) 3.Foster, I., Kesselman, C.: The Grid: Blueprint for a New Computing Infrastructure. Mor-gan Kaufmann, San Francisco (1999)104.Czajkowski, K., Fitzgerald, S., Foster, I., Kesselman, C.: Grid Information Services forDistributed Resource Sharing. In: 10th IEEE International Symposium on High Perfor-mance Distributed Computing, pp. 181–184. IEEE Press, New York (2001)5.Foster, I., Kesselman, C., Nick, J., Tuecke, S.: The Physiology of the Grid: an Open GridServices Architecture for Distributed Systems Integration. Technical report, Global Grid Forum (2002)6.National Center for Biotechnology Information, 。
Springer Lecture Notes in Computer Science (LNCS 1414) pp. 45-57.
A decade later we know that despite o ering certain bene t, something like the active badge location system did not become standard o ce equipment although useful, the technology did not get weaved into the fabric of o ce life or everyday life. Considering the technical progress over the past ten years, it is reasonable to assume that it is not technical problems that have prevented such systems from becoming standard o ce equipment.
发表高级别论文的心得（五篇模版）
发表高级别论文的心得（五篇模版）第一篇：发表高级别论文的心得【转载】发表高级别论文的心得（转贴）2010年01月28日星期四 07:53 P.M.发表论文的一些体会如何发表高水平论文SCI/EI/ISTP/一级期刊的基本知识;如何利用数据库和查找文献；如何寻找领域前沿；如何撰写高水平论文和投稿；把握数量和质量的平衡。
SCI索引SCI（科学引文索引，英文全称Science Citation Index）是美国科学情报研究所（Institute for Scientific Information，简称ISI）拥有的世界著名的期刊文献检索工具。
SCI是SCI的光盘核心库；而SCI-Expanded(简称SCIE)是SCI的扩展库。
国内高校在统计论文索引情况时一般对SCI和SCIE都认可为SCI索引。
ISTP会议录索引ISI Proceedings(ISTP-Index to Scientific & Technical Proceedings)，ISTP科学技术会议录索引是美国ISI编辑出版的查阅各种会议录的网络数据库。
目前国内很多高校也在统计ISTP索引的数量。
一般而言SCI索引的会议必然会被ISTP同时索引，但是反之不然。
影响因子IF 期刊引用报告JCR(Jorunal Citation Reports)是ISI对其SCI索引的期刊进行的参数化评价，影响因子IF(Impact Factor)是其中一项最有代表性的参数。
IF是当年其它SCI论文引用该刊此前2年所发表文章次数除以该刊前2中发表的文章数目，其值越大，说明该期刊越重要。
影响因子IF举例2004年计算机软件工程大类方面的76种期刊中，影响因子最大的是JOURNAL OF MACHINE LEARNING RESEARCH(5.952),最小的是COMPUTER GRAPHICS WORLD为0,国内最高水平的JOURNAL OF COMPUTER SCIENCE&TECHNOLOGY为0.28，2004年LNCS的影响因子为0.513 需要强调单篇文章的引用次数问题，自引与他引的问题。
springer-book-chapter
Lecture Notes in Computer Science: Authors’Instructions for the Preparation of Camera-Ready Contributionsto LNCS/LNAI/LNBI Proceedings Alfred Hofmann ,Ursula Barth,Ingrid Haas,Frank Holzwarth,Anna Kramer,Leonie Kunz,Christine Reiß,Nicole Sator,Erika Siebert-Cole,and Peter StraßerSpringer-Verlag,Computer Science Editorial,Tiergartenstr.17,69121Heidelberg,Germany{alfred.hofmann,ursula.barth,ingrid.haas,frank.holzwarth,anna.kramer,leonie.kunz,christine.reiss,nicole.sator,erika.siebert-cole,peter.strasser,lncs}@/lncsAbstract.The abstract should summarize the contents of the paperand should contain at least70and at most150words.It should bewritten using the abstract environment.Keywords:We would like to encourage you to list your keywords withinthe abstract section1IntroductionYou are strongly encouraged to use L A T E X2εfor the preparation of your camera-ready manuscript together with the corresponding Springer classfile llncs.cls. Only if you use L A T E X2εcan hyperlinks be generated in the online version of your manuscript.The L A T E X source of this instructionfile for L A T E X users may be used as a tem-plate.This is located in the“authors”subdirectory in ftp://ftp.springer.de/ pub/tex/latex/llncs/latex2e/instruct/and entitled typeinst.tex.There is a separate package for Word users.Kindly send thefinal and checked source and PDFfiles of your paper to the Contact Volume Editor.This is usually one of the organizers of the conference.You should make sure that the L A T E X and the PDFfiles are identical and correct and that only one version of your paper is sent.It is not possible to updatefiles at a later stage.Please note that we do not need the printed paper.Please note that the LNCS Editorial assumes that all authors have used the west-ern naming convention,with given names preceding surnames.This determines the structure of the names in the running heads and the author index.2Lecture Notes in Computer Science:Authors’InstructionsWe would like to draw your attention to the fact that it is not possible to modify a paper in any way,once it has been published.This applies to both the printed book and the online version of the publication.Every detail,including the order of the names of the authors,should be checked before the paper is sentto the Volume Editors.1.1Checking the PDF FileKindly assure that the Contact Volume Editor is given the name and email address of the contact author for your paper.The Contact Volume Editor uses these details to compile a list for our production department at SPS in India. Once thefiles have been worked upon,SPS sends a copy of thefinal pdf of each paper to its contact author.The contact author is asked to check through thefinal pdf to make sure that no errors have crept in during the transfer or preparation of thefiles.This should not be seen as an opportunity to update or copyedit the papers,which is not possible due to time constraints.Only errors introduced during the preparation of thefiles will be corrected.This round of checking takes place about two weeks after thefiles have been sent to the Editorial by the Contact Volume Editor,i.e.,roughly seven weeks before the start of the conference for conference proceedings,or seven weeks before the volume leaves the printer’s,for post-proceedings.If SPS does not receive a reply from a particular contact author,within the timeframe given, then it is presumed that the author has found no errors in the paper.The tight publication schedule of LNCS does not allow SPS to send reminders or search for alternative email addresses on the Internet.In some cases,it is the Contact Volume Editor that checks all thefinal pdfs.In such cases,the authors are not involved in the checking phase.1.2Additional Information Required by the Volume EditorIf you have more than one surname,please make sure that the Volume Editor knows how you are to be listed in the author index.1.3Copyright FormsThe copyright form may be downloaded from the“For Authors”(Information for LNCS Authors)section of the LNCS Website:/lncs.Please send your signed copyright form to the Contact Volume Editor,either as a scanned pdf or by fax or by courier.One author may sign on behalf of all of the other authors of a particular paper.Digital signatures are acceptable.2Paper PreparationSpringer provides you with a complete integrated L A T E X document class(llncs.cls) for multi-author books such as those in the LNCS series.Papers not complyingLecture Notes in Computer Science:Authors’Instructions3 with the LNCS style will be reformatted.This can lead to an increase in the overall number of pages.We would therefore urge you not to squash your paper.Please always cancel any superfluous definitions that are not actually used in your text.If you do not,these may conflict with the definitions of the macro package,causing changes in the structure of the text and leading to numerous mistakes in the proofs.If you wonder what L A T E X is and where it can be obtained,see the“La-TeX project site”()and especially the web-page“How to get it”(/ftp.html)respectively.When you use L A T E X together with our document classfile,llncs.cls,your text is typeset automatically in Computer Modern Roman(CM)fonts.Please do not change the preset fonts.If you have to use fonts other than the preset fonts,kindly submit these with yourfiles.Please use the commands\label and\ref for cross-references and the com-mands\bibitem and\cite for references to the bibliography,to enable us to create hyperlinks at these places.For preparing yourfigures electronically and integrating them into your sourcefile we recommend using the standard L A T E X graphics or graphicx pack-age.These provide the\includegraphics command.In general,please refrain from using the\special command.Remember to submit any further stylefiles and fonts you have used together with your sourcefiles.Headings.Headings should be capitalized(i.e.,nouns,verbs,and all other words except articles,prepositions,and conjunctions should be set with an initial capital)and should,with the exception of the title,be aligned to the left.Words joined by a hyphen are subject to a special rule.If thefirst word can stand alone, the second word should be capitalized.Here are some examples of headings:“Criteria to Disprove Context-Freeness of Collage Language”,“On Correcting the Intrusion of Tracing Non-deterministic Programs by Software”,“A User-Friendly and Extendable Data Distribution System”,“Multi-flip Networks:Parallelizing GenSAT”,“Self-determinations of Man”.Lemmas,Propositions,and Theorems.The numbers accorded to lemmas, propositions,and theorems,etc.should appear in consecutive order,starting with Lemma1,and not,for example,with Lemma11.2.1FiguresFor L A T E X users,we recommend using the graphics or graphicx package and the \includegraphics command.Please check that the lines in line drawings are not interrupted and are of a constant width.Grids and details within thefigures must be clearly legible and may not be written one on top of the other.Line drawings should have4Lecture Notes in Computer Science:Authors’Instructionsa resolution of at least800dpi(preferably1200dpi).The lettering infigures should have a height of2mm(10-point type).Figures should be numbered and should have a caption which should always be positioned under thefigures,in contrast to the caption belonging to a table,which should always appear above the table;this is simply achieved as matter of sequence in your source.Please center thefigures or your tabular material by using the\centering declaration.Short captions are centered by default between the margins and typeset in9-point type(Fig.1shows an example).The distance between text andfigure is preset to be about8mm,the distance betweenfigure and caption about6mm.To ensure that the reproduction of your illustrations is of a reasonable quality, we advise against the use of shading.The contrast should be as pronounced as possible.If screenshots are necessary,please make sure that you are happy with the print quality before you send thefiles.Fig.1.One kernel at x s(dotted kernel)or two kernels at x i and x j(left and right) lead to the same summed estimate at x s.This shows afigure consisting of different types of lines.Elements of thefigure described in the caption should be set in italics, in parentheses,as shown in this sample caption.Please definefigures(and tables)asfloating objects.Please avoid using op-tional location parameters like“[h]”for“here”.Remark1.In the printed volumes,illustrations are generally black and white (halftones),and only in exceptional cases,and if the author is prepared to cover the extra cost for color reproduction,are colored pictures accepted.Colored pictures are welcome in the electronic version free of charge.If you send colored figures that are to be printed in black and white,please make sure that theyLecture Notes in Computer Science:Authors’Instructions5 really are legible in black and white.Some colors as well as the contrast of converted colors show up very poorly when printed in black and white.2.2FormulasDisplayed equations or formulas are centered and set on a separate line(with an extra line or halfline space above and below).Displayed expressions should be numbered for reference.The numbers should be consecutive within each section or within the contribution,with numbers enclosed in parentheses and set on the right margin–which is the default if you use the equation environment,e.g.,ψ(u)=To12Λ−1o u,u+N∗(−u)dt.(1)Equations should be punctuated in the same way as ordinary text but witha small space before the end punctuation mark.2.3FootnotesThe superscript numeral used to refer to a footnote appears in the text either directly after the word to be discussed or–in relation to a phrase or a sentence–following the punctuation sign(comma,semicolon,or period).Footnotes should appear at the bottom of the normal text area,with a line of about2cm set immediately above them.12.4Program CodeProgram listings or program commands in the text are normally set in typewriter font,e.g.,CMTT10or Courier.Example of a Computer Programprogram Inflation(Output){Assuming annual inflation rates of7%,8%,and10%,...years};constMaxYears=10;varYear:0..MaxYears;Factor1,Factor2,Factor3:Real;beginYear:=0;Factor1:=1.0;Factor2:=1.0;Factor3:=1.0;WriteLn(’Year7%8%10%’);WriteLn;1The footnote numeral is setflush left and the text follows with the usual word spacing.6Lecture Notes in Computer Science:Authors’InstructionsrepeatYear:=Year+1;Factor1:=Factor1*1.07;Factor2:=Factor2*1.08;Factor3:=Factor3*1.10;WriteLn(Year:5,Factor1:7:3,Factor2:7:3,Factor3:7:3)until Year=MaxYearsend.(Example from Jensen K.,Wirth N.(1991)Pascal user manual and report.Springer, New York)2.5CitationsFor citations in the text please use square brackets and consecutive numbers:[1],[2],[4]–provided automatically by L A T E X’s\cite...\bibitem mechanism.2.6Page Numbering and Running HeadsThere is no need to include page numbers.If your paper title is too long to serve as a running head,it will be shortened.Your suggestion as to how to shorten it would be most welcome.3LNCS OnlineThe online version of the volume will be available in LNCS Online.Members of institutes subscribing to the Lecture Notes in Computer Science series have access to all the pdfs of all the online publications.Non-subscribers can only read as far as the abstracts.If they try to go beyond this point,they are automatically asked,whether they would like to order the pdf,and are given instructions as to how to do so.Please note that,if your email address is given in your paper,it will also be included in the meta data of the online version.4BibTeX EntriesThe correct BibTeX entries for the Lecture Notes in Computer Science volumes can be found at the following Website shortly after the publication of the book: rmatik.uni-trier.de/~ley/db/journals/lncs.html Acknowledgments.The heading should be treated as a subsubsection heading and should not be assigned a number.Lecture Notes in Computer Science:Authors’Instructions7 5The References SectionIn order to permit cross referencing within LNCS-Online,and eventually between different publishers and their online databases,LNCS will,from now on,be standardizing the format of the references.This new feature will increase the visibility of publications and facilitate academic research considerably.Please base your references on the examples below.References that don’t adhere to this style will be reformatted by Springer.You should therefore check your references thoroughly when you receive thefinal pdf of your paper.The reference section must be complete.You may not omit references.Instructions as to where tofind a fuller version of the references are not permissible.We only accept references written using the latin alphabet.If the title of the book you are referring to is in Russian or Chinese,then please write(in Russian) or(in Chinese)at the end of the transcript or translation of the title.The following section shows a sample reference list with entries for journal articles[1],an LNCS chapter[2],a book[3],proceedings without editors[4]and [5],as well as a URL[6].Please note that proceedings published in LNCS are not cited with their full titles,but with their acronyms!References1.Smith,T.F.,Waterman,M.S.:Identification of Common Molecular Subsequences.J.Mol.Biol.147,195–197(1981)2.May,P.,Ehrlich,H.C.,Steinke,T.:ZIB Structure Prediction Pipeline:Composinga Complex Biological Workflow through Web Services.In:Nagel,W.E.,Walter,W.V.,Lehner,W.(eds.)Euro-Par2006.LNCS,vol.4128,pp.1148–1158.Springer, Heidelberg(2006)3.Foster,I.,Kesselman,C.:The Grid:Blueprint for a New Computing Infrastructure.Morgan Kaufmann,San Francisco(1999)4.Czajkowski,K.,Fitzgerald,S.,Foster,I.,Kesselman,C.:Grid Information Servicesfor Distributed Resource Sharing.In:10th IEEE International Symposium on High Performance Distributed Computing,pp.181–184.IEEE Press,New York(2001) 5.Foster,I.,Kesselman,C.,Nick,J.,Tuecke,S.:The Physiology of the Grid:an OpenGrid Services Architecture for Distributed Systems Integration.Technical report, Global Grid Forum(2002)6.National Center for Biotechnology Information, Appendix:Springer-Author DiscountLNCS authors are entitled to a33.3%discount offall Springer publications. Before placing an order,the author should send an email,giving full details of his or her Springer publication,to orders-HD-individuals@ to obtain a so-called token.This token is a number,which must be entered when placing an order via the Internet,in order to obtain the discount.8Lecture Notes in Computer Science:Authors’Instructions6Checklist of Items to be Sent to Volume EditorsHere is a checklist of everything the volume editor requires from you:Thefinal L A T E X sourcefilesAfinal PDFfileA copyright form,signed by one author on behalf of all of the authors of thepaper.A readme giving the name and email address of the corresponding author.。
学术英语课后答案 unit1
学术英语理工教师手册Unit 1 Choosing a TopicI Teaching ObjectivesIn this unit , you will learn how to:1.choose a particular topic for your research2.formulate a research question3.write a working title for your research essay4.enhance your language skills related with reading and listening materials presented in this unit II. Teaching Procedures1.Deciding on a topicTask 1Answers may vary.Task 21 No, because they all seem like a subject rather than a topic, a subject which cannot be addressed even by a whole book, let alone by a1500-wordessay.2Each of them can be broken down into various and more specific aspects. For example, cancer can be classified into breast cancer, lung cancer, liver cancer and so on. Breast cancer can have such specific topics for research as causes for breast cancer, effects of breast cancer and prevention or diagnosis of breast cancer.3 Actually the topics of each field are endless. Take breast cancer for example, we can have the topics like:Why Women Suffer from Breast Cancer More Than Men?A New Way to Find Breast TumorsSome Risks of Getting Breast Cancer in Daily LifeBreast Cancer and Its Direct Biological ImpactBreast Cancer—the Symptoms & DiagnosisBreastfeeding and Breast CancerTask 31 Text 1 illustrates how hackers or unauthorized users use one way or another to get inside a computer, while Text2 describes the various electronic threats a computer may face.2 Both focus on the vulnerability of a computer.3 Text 1 analyzes the ways of computer hackers, while Text 2 describes security problems of a computer.4 Text 1: The way hackers “get inside” a computerText 2: Electronic threats a computer facesYes, I think they are interesting, important, manageable and adequate.Task 41Lecture1:Ten Commandments of Computer EthicsLecture 2:How to Deal with Computer HackersLecture 3:How I Begin to Develop Computer Applications2Answersmay vary.Task 5Answers may vary.2 Formulating a research questionTask 1Text 3Research question 1: How many types of cloud services are there and what are they? Research question 2: What is green computing?Research question 3: What are advantages of the cloud computing?Text 4Research question 1: What is the Web 3.0?Research question 2: What are advantages and disadvantages of the cloud computing? Research question 3: What security benefits can the cloud computing provide?Task 22 Topic2: Threats of Artificial IntelligenceResearch questions:1) What are the threats of artificial intelligence?2) How can human beings control those threats?3) What are the difficulties to control those threats?3 Topic3: The Potentials of NanotechnologyResearch questions:1) What are its potentials in medicine?2) What are its potentials in space exploration?3) What are its potentials in communications?4 Topic4: Global Warming and Its EffectsResearch questions:1) How does it affect the pattern of climates?2) How does it affect economic activities?3) How does it affect human behavior?Task 3Answers may vary.3 Writing a working titleTask 1Answers may vary.Task 21 Lecture 4 is about the security problems of cloud computing, while Lecture 5 is about the definition and nature of cloud computing, hence it is more elementary than Lecture 4.2 The four all focus on cloud computing. Although Lecture 4 and Text 4 address the same topic, the former is less optimistic while the latter has more confidence in the security of cloud computing. Text3 illustrates the various advantages of cloud computing.3 Lecture 4: Cloud Computing SecurityLecture 5: What Is Cloud Computing?Task 3Answers may vary.4 Enhancing your academic languageReading: Text 11.Match the words with their definitions.1g 2a 3e 4b 5c 6d 7j 8f 9h 10i2. Complete the following expressions or sentences by using the target words listed below with the help of the Chinese in brackets. Change the form if necessary.1 symbolic 2distributed 3site 4complex 5identify6fairly 7straightforward 8capability 9target 10attempt11process 12parameter 13interpretation 14technical15range 16exploit 17networking 18involve19 instance 20specification 21accompany 22predictable 23profile3. Read the sentences in the box. Pay attention to the parts in bold.Now complete the paragraph by translating the Chinese in brackets. You may refer to the expressions and the sentence patterns listed above.ranging from（从……到）arise from some misunderstandings（来自于对……误解）leaves a lot of problems unsolved（留下很多问题没有得到解决）opens a path for（打开了通道）requires a different frame of mind（需要有新的思想）4.Translate the following sentences from Text 1 into Chinese.1) 有些人声称黑客是那些超越知识疆界而不造成危害的好人（或即使造成危害，但并非故意而为），而“骇客”才是真正的坏人。
Springer-word模板
Lecture Notes in Computer Science: Authors’ Instructions for the Preparationof Camera-Ready Contributionsto LNCS/LNAI/LNBI ProceedingsAlfred Hofmann1,1, Brigitte Apfel1, Ursula Barth1, ChristineGünther1,Ingrid Haas1, Frank Holzwarth1, Anna Kramer1, Leonie Kunz1,Nicole Sator1, Erika Siebert-Cole1 and Peter Straßer1,1 Springer-Verlag, Computer Science Editorial, Tiergartenstr. 17,69121 Heidelberg, Germany{Alfred.Hofmann, Brigitte.Apfel, Ursula.Barth, Christine.Guenther, Ingrid.Haas, Frank.Holzwarth, Anna.Kramer, Leonie.Kunz, Nicole.Sator, Erika.Siebert-Cole, Peter.Strasser, LNCS}Springer.Abstract.The abstract should summarize the contents of thepaper and should contain at least 70 and at most 150 words. Itshould be set in 9-point font size and should be inset 1.0 cmfrom the right and left margins. There should be two blank (10-point) lines before and after the abstract. This document is inthe required format.Keywords: We would like to encourage you to list your keywordsin this section.1 IntroductionThis instruction file for Word users (there is a separate instruction file for LaTeX users) may be used as a template. Kindly send the final and checked Word and PDF files of your paper to the Contact Volume Editor. This is usually one of the organizers of the conference. You should make sure that the Word and the PDF files are identical and correct and that only one version of your paper is sent.1Please note that the LNCS Editorial assumes that all authors have used the westernnaming convention, with given names preceding surnames. This determines thestructure of the names in the running heads and the author index.It is not possible to update files at a later stage. Please note that we do not need the printed paper.We would like to draw your attention to the fact that it is not possible to modify a paper in any way, once it has been published. This applies to both the printed book and the online version of the publication. Every detail, including the order of the names of the authors, should be checked before the paper is sent to the Volume Editors.1.1 Checking the PDF FileKindly assure that the Contact Volume Editor is given the name and email address of the contact author for your paper. The Contact Volume Editor uses these details to compile a list for our production department at SPS in India. Once the files have been worked upon, SPS sends a copy of the final pdf of each paper to its contact author. The contact author is asked to check through the final pdf to make sure that no errors have crept in during the transfer or preparation of the files. This should not be seen as an opportunity to update or copyedit the papers, which is not possible due to time constraints. Only errors introduced during the preparation of the files will be corrected.This round of checking takes place about two weeks after the files have been sent to the Editorial by the Contact Volume Editor, i.e., roughly seven weeks before the start of the conference for conference proceedings, or seven weeks before the volume leaves the printer’s, for post-proceedings. If SPS does not receive a reply from a particular contact author, within the timeframe given, then it is presumed that the author has found no errors in the paper. The tight publication schedule of LNCS does not allow SPS to send reminders or search for alternative email addresses on the Internet.In some cases, it is the Contact Volume Editor that checks all the pdfs. In such cases, the authors are not involved in the checking phase.1.2 Additional Information Required by the Volume EditorIf you have more than one surname, please make sure that the Volume Editor knows how you are to be listed in the author index.1.3 Copyright FormsThe copyright form may be downloaded from the For Authors section of the LNCS Webpage: .springer./lncs. Please send your signed copyright form to the Contact Volume Editor, either as a scanned pdf or by fax or by courier. One author may sign on behalf of all of the other authors of a particular paper. Digital signatures are acceptable.2 Paper PreparationThe printing area is 122 mm × 193 mm. The text should be justified to occupy the full line width, so that the right margin is not ragged, with words hyphenated as appropriate. Please fill pages so that the length of the text is no less than 180 mm, if possible.Use 10-point type for the name(s) of the author(s) and 9-point type for the address(es) and the abstract. For the main text, please use 10-point type and single-line spacing. We recommend the use of Computer Modern Roman or Times. Italic type may be used to emphasize words in running text. Bold type and underlining should be avoided.Papers not complying with the LNCS style will be reformatted. This can lead to an increase in the overall number of pages. We would therefore urge you not to squash your paper.Headings.Headings should be capitalized (i.e., nouns, verbs, and all other words except articles, prepositions, and conjunctions should be set with an initial capital) and should, with the exception of the title, be aligned to the left. Words joined by a hyphen are subject to a special rule. If the first word can stand alone, the second word should be capitalized. The font sizes are given in Table 1.Here are some examples of headings: "Criteria to Disprove Context-Freeness of Collage Languages", "On Correcting the Intrusion of Tracing Non-deterministic Programs by Software", "A User-Friendly and Extendable Data Distribution System", "Multi-flip Networks: Parallelizing GenSAT", "Self-determinations of Man".Table 1. Font sizes of headings. Table captions should always be positioned above the tables.Heading level Example Font size andstyleTitle (centered) Lecture Notes …14 point, bold1st-level heading 1 Introduction12 point, bold2nd-level heading 2.1 Printing Area10 point, bold10 point, bold3rd-level heading Headings.Textfol lows …4th-level heading Remark. Text f ollows …10 point, italicLemmas, Propositions, and Theorems.The numbers accorded to lemmas, propositions, and theorems, etc. should appear in consecutive order, starting with Lemma 1, and not, for example, with Lemma 11.2.1 FiguresPlease check that the lines in line drawings are not interrupted and have a constant width. Grids and details within the figures must be clearly legible and may not be written one on top of the other. Line drawings should have a resolution of at least 800 dpi (preferably 1200 dpi). The lettering in figures should have a height of 2 mm (10-point type). Figures should be numbered and should have a caption which should always be positioned under the figures, in contrast to the caption belonging to a table, which should always appear above the table. Please center the captions between the margins and set them in 9-point type (Fig. 1 shows an example). The distance between text and figure should be about 8 mm, the distance between figure and caption about 6 mm.To ensure that the reproduction of your illustrations is of a reasonable quality, we advise against the use of shading. The contrast should be as pronounced as possible.If screenshots are necessary, please make sure that you are happy with the print quality before you send the files.Remark 1. In the printed volumes, illustrations are generally black and white (halftones), and only in exceptional cases, and if the author is prepared to cover the extra costs involved, are colored pictures accepted. Colored pictures are welcome in the electronic version free of charge. If you send colored figures that are to be printed in black and white, please make sure that they really are legible in black and white. Some colors show up very poorly when printed in black and white.Fig. 1. One kernel at x s (dotted kernel) or two kernels at x i and x j (left and right) lead to the same summed estimate at x s. This shows a figure consisting of different types of lines. Elements of the figure described in the caption should be set in italics, in parentheses, as shown in this sample caption.2.2 FormulasDisplayed equations or formulas are centered and set on a separate line (with an extra line or halfline space above and below). Displayed expressions should be numbered for reference. The numbers should be consecutive within each section or within the contribution, with numbers enclosed in parentheses and set on the right margin.x + y = z . (1)Equations should be punctuated in the same way as ordinary text but with a small space before the end punctuation mark.2.3 FootnotesThe superscript numeral used to refer to a footnote appears in the text either directly after the word to be discussed or – in relation to a phrase or a sentence – following the punctuation mark (comma, semicolon, or period). Footnotes should appear at the bottom of the normal text area, with a line of about 5cm set immediately above them2.2.4 Program CodeProgram listings or program commands in the text are normally set in typewriter font, e.g., CMTT10 or Courier.Example of a Computer Program from Jensen K., Wirth N. (1991) Pascal user manual and report. Springer, New Yorkprogram Inflation (Output){Assuming annual inflation rates of 7%, 8%, and10%,... years};const MaxYears = 10;var Year: 0..MaxYears;Factor1, Factor2, Factor3: Real;beginYear := 0;Factor1 := 1.0; Factor2 := 1.0; Factor3 := 1.0;WriteLn('Year 7% 8% 10%'); WriteLn;repeatYear := Year + 1;Factor1 := Factor1 * 1.07;Factor2 := Factor2 * 1.08;Factor3 := Factor3 * 1.10;WriteLn(Year:5,Factor1:7:3,Factor2:7:3,Factor3:7:3)until Year = MaxYearsend.2The footnote numeral is set flush left and the text follows with the usual word spacing.2.5 CitationsFor citations in the text please use square brackets and consecutive numbers: [1], [2], [3], etc.2.6 Page Numbering and Running HeadsThere is no need to include page numbers. If your paper title is too long to serve as a running head, it will be shortened. Your suggestion as to how to shorten it would be most welcome.3 LNCS OnlineThe online version of the volume will be available in LNCS Online. Members of institutes subscribing to the Lecture Notes in Computer Science series have access to all the pdfs of all the online publications. Non-subscribers can only read as far as the abstracts. If they try to go beyond this point, they are automatically asked, whether they would like to order the pdf, and are given instructions as to how to do so.Please note that, if your email address is given in your paper, it will also be included in the meta data of the online version.4 BibTeX EntriesThe correct BibTeX entries for the Lecture Notes in Computer Science volumes can be found at the following Website shortly after the publication of the book: .informatik.uni-trier.de/~ley/db/journals/lncs.htmlAcknowledgments. The heading should be treated as a 3rd level heading and should not be assigned a number.5 The References SectionIn order to permit cross referencing within LNCS-Online, and eventually between different publishers and their online databases,LNCS will, from now on, be standardizing the format of the references. This new feature will increase the visibility of publications and facilitate academic research considerably. Please base your references on the examples below. References that don’t adhere to this style will be reformatted by Springer. You should therefore check your references thoroughly when you receive the final pdf of your paper. The reference section must be complete. You may not omit references. Instructions as to where to find a fuller version of the references are not permissible.We only accept references written using the latin alphabet. If the title of the book you are referring to is in Russian or Chinese, then please write (in Russian) or (in Chinese) at the end of the transcript or translation of the title.The following section shows a sample reference list with entries for journal articles [1], an LNCS chapter [2], a book [3], proceedings without editors [4] and [5], as well as a URL [6]. Please note that proceedings published in LNCS are not cited with their full titles, but with their acronyms!References1. Smith, T.F., Waterman, M.S.: Identification of Common MolecularSubsequences. J. Mol. Biol. 147, 195--197 (1981)2. May, P., Ehrlich, H.C., Steinke, T.: ZIB Structure Prediction Pipeline:Composing a Complex Biological Workflow through Web Services. In: Nagel, W.E., Walter, W.V., Lehner, W. (eds.) Euro-Par 2006. LNCS, vol. 4128, pp.1148--1158. Springer, Heidelberg (2006)3. Foster, I., Kesselman, C.: The Grid: Blueprint for a New ComputingInfrastructure. Morgan Kaufmann, San Francisco (1999)4. Czajkowski, K., Fitzgerald, S., Foster, I., Kesselman, C.: GridInformation Services for Distributed Resource Sharing. In: 10th IEEE International Symposium on High Performance Distributed Computing, pp.181--184. IEEE Press, New York (2001)5. Foster, I., Kesselman, C., Nick, J., Tuecke, S.: The Physiology of theGrid: an Open Grid Services Architecture for Distributed Systems Integration. Technical report, Global Grid Forum (2002)6. National Center for Biotechnology Information, Appendix: Springer-Author DiscountLNCS authors are entitled to a 33.3% discount off all Springer publications. Before placing an order, they should send an email to orders-HD-individualsspringer., giving full details of their Springer publication, to obtain a so-called token. This token is a number, which must be entered when placing an order via the Internet, in order to obtain the discount.Checklist of Items to be Sent to Volume Editors1.A final Word or RTF file2.A final PDF file3.A copyright form, signed by one author on behalf of all of theauthors of the paper4.A readme giving the name and email address of the correspondingauthor。
Lecture Notes in Computer Science 1
Lecture Notes in Computer Science 1 Multiscale feature extraction from the visualenvironment in an active vision systemY.Machrouh1, J.-S.Liénard1, P.Tarroux1,2Abstract. This paper presents a visual architecture able to identify salient re-gions in a visual scene and to use them to focus on interesting locations. It is in-spired by the ability of natural vision systems to perform a differential process-ing of spatial frequencies both in time and space and to focus their attention ona very local part of the visual scene. The present paper analyzes how this dif-ferential processing of spatial frequencies is able to provide an artificial systemwith the information required to perform an exploration of its visual worldbased on a center-surround distinction of the external scene. It shows how thesalient locations can be gathered on the basis of their similarities to form a highlevel representation of the visual scene.IntroductionThe use of active mechanisms seems to be a way to improve the abilities of machine vision systems. Active systems search salient features in the visual scene through a dynamic exploration. They can direct their search toward the most meaningful stimuli using attentional mechanisms leading to a reduction of the computational load [1,2].Thus, natural vision is a behavioral task, not a passive filtering process. An explora-tion of the visual world that relates perception and action allows to label the external space with natural landmarks associated with the exploratory behavior. In this re-spect, the relationships between agents and natural systems suggest that certain as-pects of natural perception can be successfully incorporated in artificial agents.Otherwise, during the past few years, several studies have been devoted to the under-standing of the essence of vision considered as an information processing mechanism[4]. This approach is grounded on Barlow’s proposal [5] which stated that the mainorganizational principle in visual systems is the reduction of the redundancy of the incoming stimuli.These considerations, issued form information theory, led several authors to analyze the statistical organization of natural images. They demonstrated that natural images (those which do not exhibit any specific bias in their pixel distribution) have a sta-tionary statistics and an auto-similar structure. As a consequence of these characteris-tics, their power spectra fall off as 1/f2 [8].In this context, different authors [6,14] demonstrated that a way to transform the initial redundancy was to improve the statistical independence of the image descrip-1LIMSI-CNRS BP 133 F-91403 Orsay Cedex2ENS 45 rue d’Ulm F-75230 Paris cedex 05Lecture Notes in Computer Science 2tors. According to this hypothesis, an image can be viewed as a linear superposition of several underlying independent sources.The filters that provide this statistical independence can be computed through the application of the source separation adequate algorithms (Infomax, BSS, ICA).One can show [6,14] that the optimal filters computed according to these principles are multiscale local orientation detectors similar to a Gabor wavelet basis [7]. However, although a lot of work has been devoted to the understanding of these theo-retical bases of information processing in natural visual system, few attempts have been made thus far to use these principles in artificial vision systems. Practical im-plementations impose some limitations that require to analyze what is really obtained with simplified models based on these general principles. On the other hand, no arti-ficial vision system has been designed to include both multiscale wavelet analysis and differential spatial and temporal processing of spatial frequencies. A prerequisite to the design of such a system is to be able to characterize the information obtained from a bank of wavelet filters in different frequency channels.We thus analyzed here the information issued from various combinations of high and low frequencies of statistically uncorrelated signals. Our aim was to determine how to build a multivariate representation of the scene that allows a dynamic grouping of image points on the basis of their similarities in a given context and for a given task.System overviewImage dataA set of 11 natural images selected from a larger database was used in the present study. Pictures that include too many traces of human activity (buildings, roads…) were avoided. Only images with similar initial resolution (around 256x512 pixels) were retained.Figure 1. Sample image from the set of natural images used in the present work.(original size 512x256)The images were discarded when their power spectrum did not fit the 1/f2 characteris-tics [8]. Figure 1 shows one typical example of an image used in the present study.Lecture Notes in Computer Science 3 Initial filtersA guideline for this work was to retain among the filtering characteristics of the pri-mate visual system those which can be useful for the elaboration of an artificial sys-tem of situated and active vision.Two characteristics have attracted our attention: the elimination of image redundancy in the processing steps designed to maximize the statistical independence of the scene descriptors and the differences in the processing of spatial frequencies between the center and the surround of the visual field.The visual scene was filtered by a first bank of Gabor wavelets in four spatial orienta-tions and four spatial frequencies (1/8, 1/16, 1/32, 1/64). For each initial image we got 32 resulting images (two for each quadrature pair of each of the 16 Gabor filter). This multiscale processing was implemented using a Burt pyramid according to the method proposed by Guérin-Dugué [10].For the purpose of this study and in order to obtain a complete view of what informa-tion is obtained from a detector during a systematic exploration of the visual scene, the whole scene was filtered by the entire bank of filters. In an operational system with a focal vision only a small part of these computations are needed.Simple cells – Complex cellsAn important distinction between the use of wavelets in image processing and the filtering steps in the visual system is the presence of strong non-linearities in the latter. Primary visual cortex shows several cell types according to the non linearities they implement. Simple cells (SC) perform an additive combination of their inputs. They respond to an oriented stimulus localized at the center of their receptive field. The so-called complex cells (CC), on the contrary, exhibit a kind of translational invariance and respond to a stimulus whatever its position in the receptive field of the cell.Figure 2. Effects of filtering of the statistical independance criterion. Init: Initial image, SC: Simple Cells, CC: Complex cellsLecture Notes in Computer Science 4 Other cell types (mainly in extrastriate cortex) combine these outputs in order to be sensitive to curvature and terminations (end-stop cells).To model simple cells we used additive units with a zero threshold ramp transfer function which amounts to take into account only the positive part of Gabor filters. The inhibitory part is indeed not transmitted by these cells.According to Field [5], we modeled complex cells output as the norm of quadrature pair Gabor filters. We verified that this implementation effectively leads to a reduc-tion of the redundancy for both cell types by a comparison of the kurtosis before and after filtering (Figure 2). Kurtosis is indeed a good measurement of the statistical independence of a set of detectors [9].A third type of detector with large receptive fields and designed to provide a contex-tual information will be considered in the following section.In order to build a set of higher level detectors suitable for the extraction of complex features we performed a Karhunen-Loeve transform of the outputs. A set of 1744 image patches (5x5) extracted randomly from the initial 11 natural images was used to build these spaces. We thus obtained 8 eigen-vectors at the output of simple cells and 4 eigen-vectors at the output of complex cells for each frequency band. These computations amount to a non-linear principal component projection of the initial image performed with two different types of non linearities.Global energy – Local contextAs stated above, we assumed the existence of detectors sensitive to the global energy in the different orientations. In an image region corresponding to the fovea, the sys-tem computes a global energy vector for each of the four orientations. This vector is used to build a signature that can be used to classify the region. Such an analysis provides us with contextual information [11,13]. We consider the identification of these contexts as a prerequisite for the recognition of objects. The importance of contextual information in natural systems can be deduced from the experimental observation that object recognition is effectively facilitated if the objects are viewed in congruent contexts [13].Thus, the system computes three output sets on each image: (i) an output directly issued from the Gabor filters filtered by a ramp function (SC), (ii) an output giving the local energy at the output of these filters analogous to the output of complex cells (CC) and (iii) a large field output providing contextual information.ResultsSimple cellsFor each image point the system provides a high dimensional vector made of 32 ori-entation components spread over 4 frequency bands for SC detectors and 16 orienta-tion components in 4 frequency bands for CC detectors.Although Gabor detectors maximize the statistical independence of their outputs, in practice they are not strictly independent. The analysis of these outputs through aLecture Notes in Computer Science 5 Karhunen-Loeve transform leads to a data representation basis that sorts the represen-tations according to their greatest statistical significance.The first axis corresponding to the highest eigen-value shows highly variable details from one scene to another (figure 3 left). It emphasizes details related to the structures present in the scene. This probably results from the fact that these structures are cor-related in a given scene due to the correlation induced by the presence of objects. They are uncorrelated from one scene to another because each scene has a different organization.Figure 3. Output of SC filters: projection of the output along the first (top) and the last (botttom) eigen-vector of the output spaceOn the contrary, details filtered by the axes corresponding to the lowest eigen-values (figure 3 right) are expected to weakly contribute to the total variance. They corre-spond to features most frequently observed from one image to another.Figure 4. Eigen-images from CC filters. The images are computed as the projection of the CC outputs on the eigen-vectors defining the output space of these filters. Columns range from high to low frequencies (from left to right: 1/8 to 1/64). Lines show the filter outputs along the principal components (top: highest variance, bottom : lowest variance).The same region revealed by the first projection axis (Figure 3 left)(% initial vari-ance : 29.4%) of the KL transform and the last projection axis (Figure 3 right)(% initial variance : 2.47%) shows that, while the first axis tends to reveal long edges thatLecture Notes in Computer Science 6 contribute significantly to the general structure of the objects, the last axis tends to reveal termination and curvature points that are not characteristic of the image struc-ture.We obtain a complex set of features along the different axes. The most representative of the presence of objects correspond to the first axes. On the others, features repre-senting complex combinations of stimuli frequently observed in natural images seem to be sorted according to their level of abstractness.Complex cellsThe same transform can be applied to the output of complex cells. Figure 4 shows the main axes of the KL transform following the computation of the Gabor norm for different spatial frequency bands.The projection axes (rows in the figure) extract distinct features from the initial image as well within the same frequency band (rows) as between different frequency bands (columns)(note that for instance the building vanishes in axis 3 projection. Figure 4 3rd row). These features are entirely different from those extracted by the output transform of SC.One can observe that high frequency details disappear in low frequency channels except for objects which exhibit frequency similarities (high frequency details re-peated over a large area like the building).Objects in the foreground, which are apparently characterized by low frequencies, appear in low frequency channels while they are not represented in high frequency band. Low frequency channels are able to distinguish features that have some spatial extension (the building or the foreground bushes).A comparison of the lowest frequency channels (Figure 4 right column) shows that the locations revealed on the different axes are largely uncorrelated, thus correspond-ing to different points of view on the scene.The lesser number of low frequency features (figure 4 right column) defines a small set of landmarks able to characterize the visual space and to guide exploratory sac-cades. This low-frequency information is the only one available in the periphery of the visual field.Correlation between channelsOne of the important questions raised by this analysis is how different are the indices obtained from the different frequency channels. If two channels correspond to the same combination of basic features, the corresponding eigen-vectors should be simi-lar. Thus, a measure of the similarity between the eigen-vectors in different frequency bands is given by the product of the eigen-matrices in these frequency bands. Using this method we compared the output spaces of respectively simple and complex cells for different frequency bands. We obtained strongly different results for the compari-son of output spaces in SC channels and in CC channels.For simple cells, the correlation between the axes of the spaces corresponding to different frequencies are low and distributed over the different axes (data not shown) while in complex cells the respectively high and low frequency bands exhibit simi-larities (table 1).Lecture Notes in Computer Science 7 Table 1. Analysis of the output space for CC detectors. The eigen-vectors corresponding to the same axes show a very high correlation between respectively high and low frequency channels. The cross-correlation between eigen vectors corresponding to different axes is usually low (not reprinted here)FrequenciesAxes f0/f1 f0/f2 f0/f3 f1/f2 f1/f3 f2/f3F1 0.990 0.442 0.410 0,365 0,330 0,996F2 0.997 0.517 0.501 0.507 0.486 0.997F3 0.991 0.363 0.370 0.425 0.424 0.995F4 0.994 0.656 0,653 0.641 0.630 0.996These results lead to the conclusion that the combination of simple cells outputs across the frequency bands underline uncorrelated details, whereas the outputs in high (resp. low) frequency bands correspond most frequently to similar stimuli.A pyramidal decomposition of the scene allows to combine these characteristics to identify spatial positions characterized by spectral compositions as diverse as possi-ble.This diversity seems to lead to a greater separability of these spatial positions and seems to be able to facilitate objet discrimination.Identification of global contextsCells sensitive to low frequencies have large receptive fields. However in higher layers of the visual system cell types that encode intermediate representations also exhibit larger receptive fields. They combine the output of the cells in the preceding layers and gather the information coming from brighter regions of the visual field.A vector that combines the global energy components associated with each frequency channel provides a suitable code for representing the whole fovea. It has been shown that such vectors can be used to classify visual scenes according to the context they belong to [11,13]. In the present study, we build such detectors in computing the mean energy provided by the output of CC cells in the four frequency bands already mentioned.To determine how spatial indices provided by the channels previously described can be used for the identification of interesting locations in the scene, we performed the following experiment:A set of salient locations are computed from the eigen-images defined previously. Points in the image are selected at random or on the basis of these salient locations. At each point the mean energies of the CC outputs in an image window correspond-ing to the fovea were computed for each frequency. We thus obtained an energy vec-tor for each of the selected point. A PCA analysis was performed on this set of vec-tors. One should keep in mind that this use of PCA differs from its use in the previous sections. The Karhunen-Loeve transform was previously used as a self-organization tool leading to a set of linear combination defining complex features frequently oc-curring in natural images. In this section, PCA should be considered as a mean to analyze the structure of the space at the output of the SC and CC filters.Lecture Notes in Computer Science 8acFigure 5. Clustering of fixation points corresponding to different regions of the visual scene. Clusters were identified on the first three principal components and the fixation points corre-sponding to each cluster plotted on the diagrams at their position in the initial image (a). (b) fixation points obtained from the second eigen-image and the second frequency channel shown Fig. 4. The other diagrams show the location of some clusters gathering salient points on the basis of their spatial frequencies and orientation properties: (c) trees and bushes, (d) building, (e) strong curvature at the border between hill and sky (f) another region of interest at the same limitWhen the locations in the image are selected at random no obvious structure were observed in the PCA space. On the contrary, when they are selected on the basis of their saliencies, clusters were identified in the PCA space. Figure 5 show the loca-tions of some of these clusters on the original image. Points corresponding to a simi-lar context are grouped into the same cluster. The example shows for instance the ability of the method to separate fixation points on the basis of their natural or artifi-cial nature (Figure 5 c and d).It should be noted that Figure 5 shows only a small sample of the structures that can be identified. Only 1/16 of the available dimensions is presented here. Thus, the method transforms the initial image into a huge set of clusters each characterized by similar spectral signatures.Discussion and conclusionThe visual filter system proposed in the present work produces a set of features that can be used to guide the exploration of the external scene. The features extracted by the non linear combination of SC channels seem rather suitable for object recognition. Features obtained from the computation of local energy (CC channels) allow a parti-tion of the image into salient regions arranged according to their frequency composi-tion. The computation of the global energy provides local context information and can be used to segment the scene on the basis of its spectral characteristics.Lecture Notes in Computer Science 9 Thus, the output of this filtering system provides on one hand locations of interest able to guide an attentional system and on the other hand clusters of locations ar-ranged according to their spectral signature.This approach can be considered as an extension of textures segmentation methods [3] to the question of the identification of contexts and an extension of the method proposed by Hérault [11] to the analysis of local contexts. However it emphasizes the relativity of the context notion; the segmentation of the visual scene in (i) a global context and (ii) objects is an oversimplificationThe visual scene is thus scattered into a set of projections on several disjoint sub-spaces. In each of these subspaces, salient points form clusters according to their similarities. These salient points are projected into disjoint sets of clusters and the corresponding objects can thus be grouped according to different points of view.An object class is not characterized by a unique high level representation, but by the transient association of a subset of properties. This association can thus dynamically depend on the current task. Objects are not considered as similar and grouped on the basis of their intrinsic properties but according to those of their properties linked to a given goal.A further step in this work will be to demonstrate how such coding abilities could indeed facilitate object classification. This requires to incorporate the present algo-rithms in the control architecture of a perceptive agent such that it can build a hierar-chy of perception-action links based on the dynamic grouping of the perceived fea-tures.ACKNOWLEDGMENTSThis work was supported by a grant from the “GIS Sciences de la cognition” CNRS.REFERENCES[1] Allport, A., Visual attention. In M.I. Posner (Ed.), Foundations of cognitive science,The MIT Press, 1989.[2] Aloimonos, Y. (Ed.), Active Perception, Lawrence Erlbaum, Hillsdale,NJ, 1993.[3] Andrey, P. and Tarroux, P., Unsupervised segmentation of Markov Random Fieldmodeled textured images using selectionist relaxation, IEEE Transactions on PatternAnalysis and Machine Intelligence, 20 (1996) 252-263.[4] Atick, J.J. and Redlich, A.N., Towards a Theory of Early Visual Processing, NeuralComputation, 2 (1990) 308-320.[5] Barlow, H.B., Possible principles underlying the transformation of sensory messages.In W. Rosenblith (Ed.), Sensory Communication, The MIT Press, cambridge, MA,1961, pp. 217-234.[6] Bell, A.J. and Sejnowski, T.J., The ''independent components'' of natural scenes areedge filters, Vision Research, 37 (1997) 3327-3338.[7] Daugman, J. and Downing, C., Gabor wavelets for statistical pattern recognition. InM.A. Arbib (Ed.), The Handbook of Brain Theory and Neural Networks, The MITPress, Cambridge, MA, 1995, pp. 414-420.Lecture Notes in Computer Science 10 [8] Field, D.J., Relations between the statistics of natural images and the response prop-erties of cortical cells, Journal of the Optical Society of America A, 4 (1987) 2379-2394.[9] Field, D.J., What is the goal of sensory coding?, Neural Computation, 6 (1994) 559-601.[10] Guérin-Dugué, A. and Palagi, P.M., Implantations de filtres de Gabor par pyramided'images passe-bas, Traitement du signal, 13 (1996) 1-11.[11] Hérault, J., Oliva, A. and Guérin-Dugué, A., Scene categorisation by curvilinearcomponent analysis of low frequency spectra. , ESANN'97, Bruges, 1997, pp. 91-96.[12] Linsker, R., Self-organization in a perceptual network, Computer Magazine, 21(1988) 105-117.[13] Oliva, A. and Schyns, P.G., Coarse blobs or fine edges? Evidence that informationdiagnosticity changes the perception of complex visual stimuli, Cognitive Psychol-ogy, 34 (1997) 72-107.[14] Olshausen, B.A. and Field, D.J., Emergence of simple-cell receptive field propertiesby learning a sparse code for natural images, Nature, 381 (1996) 607-609.。
Electronic Notes in Theoretical Computer Science
DCM2005Preliminary VersionAbstract Effective ModelsUdi Boker1,2School of Computer ScienceTel Aviv UniversityTel Aviv69978,IsraelNachum Dershowitz3School of Computer ScienceTel Aviv UniversityTel Aviv69978,IsraelAbstractWe modify Gurevich’s notion of abstract machine so as to encompass computational models,that is,sets of machines that share the same domain.We also add an effec-tiveness requirement.The resultant class of“Effective Models”includes all known Turing-complete state-transition models,operating over any countable domain.Key words:Computational models,Turing machines,ASM,Abstract State Machines,Effectiveness1Sequential ProceduresWefirst define“sequential procedures”,along the lines of the“sequential algorithms”of[3].These are abstract state transition systems,whose states are algebras.Definition1.1(States)•A state is a structure(algebra)s over a(finite-arity)vocabulary F,that is, a domain(nonempty set of elements)D together with interpretations[[f]]s over D of the function names f∈F.•A location of vocabulary F over a domain D is a pair,denoted f(a),where f is a k-ary function name in F and a∈D k.1This work was carried out in partial fulfillment of the requirements for the Ph.D.degree of thefirst author.2Email:udiboker@tau.ac.il3Email:nachumd@tau.ac.ilThis is a preliminary version.Thefinal version will be published inElectronic Notes in Theoretical Computer ScienceURL:www.elsevier.nl/locate/entcs•The value of a location f(a)in a state s,denoted[[f(a)]]s,is the domain element[[f]]s(a).•We sometimes use a term f(t1,...,t k)to refer to the location f([[t1]]s,...,[[t k]]s).•Two states s and s over vocabulary F with the same domain coincide over a set T of F-terms if[[t]]s=[[t]]s for all terms t∈T.•An update of location l over domain D is a pair,denoted l:=v,where v∈D.•The modification of a state s into another state s over the same vocabulary and domain is∆(s,s )={l:=v |[[l]]s=[[l]]s =v }.•A mappingρ(s)of state s over vocabulary F and domain D via injection ρ:D→D is a state s of vocabulary F over D ,such thatρ([[f(a)]]s)= [[f(ρ(a))]]s for every location f(a)of s.•Two states s and s over the same vocabulary with domains D and D , respectively,are isomorphic if there is a bijectionπ:D↔D ,such that s =π(s).A“sequential procedure”is like Gurevich’s[3]“sequential algorithm”,with two modifications for computing a specific function,rather than expressing an abstract algorithm:the procedure vocabulary includes special constants“In”and“Out”;there is a single initial state,up to changes in In.Definition1.2(Sequential Procedures)•A sequential procedure A is a tuple F,In,Out,D,S,S0,τ ,where:F is afinite vocabulary;In and Out are nullary function names in F;D,the procedure domain,is a domain;S,its states,is a collection of structures of vocabulary F,closed under isomorphism;S0,the initial states,is a subset of S over the domain D,containing equal states up to changes in the value of In(often referred to as a single state s0);andτ:S→S,the transition function,such that:·Domain invariance.The domain of s andτ(s)is the same for every state s∈S.·Isomorphism preservation.The transition function preserves isomor-phism.Meaning,if states s and s are isomorphic via a bijectionπ,then τ(s)andτ(s )are also isomorphic viaπ.That is,τ(π(s))=π(τ(s)).·Bounded exploration.There exists afinite set T of“critical”terms, such that∆(s,τ(s))=∆(s ,τ(s ))if s and s coincide over T.Tuple elements of a procedure A are indexed F A,τA,etc.•A run of a procedure A is afinite or infinite sequence s0;τs1;τs2;τ···,where s0is an initial state and every s i+1=τA(s i).•A run s0;τs1;τs2;τ···terminates if it isfinite or if s i=s i+1from some point on.•The terminating state of a terminating run s0;τs1;τs2;τ···is2its last state if it is finite,or its stable state if it is infinite.If there is a terminating run beginning with state s and terminating in state s ,we write s ;!τs .•The extensionality of a sequential procedure A over domain D is the partial function f :D →D ,such that f (x )=[[Out ]]s whenever there’s a run s ;!τs with [[In ]]s =x ,and is undefined otherwise.Domain invariance simply ensures that a specific “run”of the procedure is over a specific domain.The isomorphism preservation reflects the fact that we are working at a fixed level of abstraction.See [3,p.89].The bounded-exploration constraint is required to ensure that the behavior of the procedure is effective.This reflects the informal assumption that the program of an algorithm can be given by a finite text [3,p.90].2Programmable MachinesThe transition function of a “programmable machine”is given by a finite “flat program”:Definition 2.1(Programmable Machines)•A flat program P of vocabulary F has the following syntax:if x 11.=y 11and x 12.=y 12and ...x 1k 1.=y 1k 1then l 1:=v 1if x 21.=y 21and x 22.=y 22and ...x 2k 2.=y 2k 2then l 2:=v 2...if x n 1.=y n 1and x n 2.=y n 2and ...x nk n .=y nk nthen l n :=v n where each .=is either ‘=’or ‘=’,n,k 1,...,k n ∈N ,and all the x ij ,y ij ,l i ,and v i are F -terms.•Each line of the program is called a rule .•The activation of a flat program P on an F -structure s ,denoted P (s ),is a set of updates {l :=v |if p then l :=v ∈P,[[p ]]s }(under the standard interpretation of =,=,and conjunction),or the empty set ∅if the above set includes two values for the same location.•A programmable machine is a tuple F ,In ,Out ,D,S ,S 0,P ,where all but the last component is as in a sequential procedure (Definition 1.2),and P is a flat program of F .•The run of a programmable machine and its extensionality are defined as for sequential procedures (Definition 1.2),where the transition function τis given by τ(s )=s ∈S such that ∆(s,s )=P (s ).To make flat programs more readable,we combine rules,as in3%commentif cond-1stat-1stat-2elsestat-3Analogous to the the main lemma of[3],one can show that every program-mable machine is a sequential procedure,and every sequential procedure is a programmable machine.In contradistinction to Abstract Sequential Machines(ASMs),we do not have built in equality,booleans,or an undefined in the definition of procedures: The equality notion is not presumed in the procedure’s initial state,nor can it be a part of the initial state of an“effective procedure”,as defined below. Rather,the transition function must be programmed to perform any needed equality checks.Boolean constants and connectives may be defined like any other constant or function.Instead of a special term for undefined values,a default domain value may be used explicitly.3Effective ModelsWe define an“effective procedure”as a sequential procedure satisfying an “initial-data”postulate(Axiom3.3below).This postulate states that the procedures may have onlyfinite initial data in addition to the domain repre-sentation(“base structure”).An“effective model”is,then,any set of effective procedures that share the same domain representation.We formalize thefiniteness of the initial data by allowing the initial state to contain an“almost-constant structure”.Since we are heading for a char-acterization of effectiveness,the domain over which the procedure actually operates should have countably many elements,which have to be nameable. Hence,without loss of generality,one may assume that naming is via terms. Definition3.1(Almost-Constant and Base Structures)•A structure S is almost constant if all but afinite number of locations have the same value.•A structure S offinite vocabulary F over a domain D is a base structure if all the domain elements are the value of a unique F-term.That is,for every element e∈D there exists a unique F-term t such that[[t]]S=e.•A structure S of vocabulary F over domain D is the union of structures S and S of vocabularies F and F ,respectively,over D,denoted S=S S , if F=F F ,[[l]]S=[[l]]S for every location l of S ,and[[l]]S=[[l]]S for every location l of S .A base structure is isomorphic to the standard free term algebra(Herbrand universe)of its vocabulary.4Proposition3.2Let S be a base structure over vocabulary G and domain D. Then:•Vocabulary G has at least one nullary function.•Domain D is countable.•Every domain element is the value of a unique location of S.Axiom3.3(Initial Data)The procedure’s initial states consist of an infi-nite base structure and an almost-constant structure.That is,for some infinite base structure BS and almost-constant structure AS,and for every initial state s0,we have s0=BS AS {In}for some In.Definition3.4(Effective Procedures and Models)•An effective procedure A is a sequential procedure satisfying the initial-data postulate.An effective procedure is,accordingly,a tuple F,In,Out,D,S,S0,τ,BS,AS ,adding a base structure BS and an almost-constant structure AS to the sequential procedure tuple,defined in Defini-tion1.2.•An effective model E is a set of effective procedures that share the same base structure.That is,BS A=BS B for all effective procedures A,B∈E.A computational model might have some predefined complex operations,as in a RAM model with built-in integer multiplication.Viewing such a model as a sequential algorithm allows the initial state to include these complex functions as oracles[3].Since we are demanding effectiveness,we cannot allow arbitrary functions as oracles,and force the initial state to include only finite data over and above the domain representation(Axiom3.3).Hence,the view of the model at the required abstraction level is accomplished by“big steps”,which may employ complex functions,while these complex functions are implemented by afinite sequence of“small steps”behind the scenes.That is,(the extensionality of)an effective procedure may be included(as an oracle) in the initial states of other effective procedures.(Cf.the“turbo”steps of[2].) 4Effective Includes ComputableTuring machines,and other computational methods,can be shown to be ef-fective.We demonstrate below how Turing machines and counter machines can be described by effective models.4.1Turing Machines.We consider Turing machines(TM)with two-way infinite tapes.The tape alphabet is{0,1}.The two edges of the tape are marked by a special$ sign.As usual,the state(instantaneous description)of a Turing machine is Left,q,Right ,where Left is afinite string containing the tape section left of the reading head,q is the internal state of the machine,and Right is afinite5string with the tape section to the right to the read head.The read head points to thefirst character of the Right string.TMs can be described by the following effective model E:Domain:Finite strings ending with a$sign.That is the domain D= {0,1}∗$.Base structure:Constructors for thefinite strings(name/arity):$/0, Cons0/1,and Cons1/1.Almost-constant structure:•Input and Output(nullary functions):In,Out.The value of In at the initial state is the content of the tape,as a string over{0,1}∗ending with a$sign.•Constants for the alphabet characters and TM-states(nullary):0,1,q0, q1,...,q k.Their initial value is irrelevant,as long it is a different value for each constant.•Variables to keep the current status of the Turing machine(nullary):Left, Right,and q.Their initial values are:Left=$,Right=$,and q=q0.•Functions to examine the tape(unary functions):Head and Tail.Their initial value,at all locations,is$.Transition function:For each Turing machine m∈TM,define an effective procedure m ∈E via aflat program looking like this:if q=q_0%TM’s state q_0if Head(Right)=0%write1,move right,switch to q_3Left:=Cons_1(Left)Right:=Tail(Right)q:=q_3%Internal operationsTail(Cons_1(Left)):=LeftHead(Cons_1(Left)):=1if Head(Right)=1%write0,move left,switch to q_1Left:=Tail(Left)Right:=Cons_0(Right)q:=q_1%Internal operationsTail(Cons_0(Right)):=RightHead(Cons_0(Left)):=0if q=q_1%TM’s state q_1...if q=q_k%the halting stateOut:=Right6The updates for Head and Tail are bookkeeping operations that are really part of the“behind-the-scenes”small steps.The procedure also requires some initialization,in order tofill the internal functions Head and Tail with their values for all strings up to the given input string.It sequentially enumerates all strings,assigning their Head and Tail values,until encountering the input string.The following internal variables (nullary functions)are used in the initialization(Name=initial value):New= $,Backward=0,Forward=1;AddDigit=0,and Direction=$.%Sequentially constructing the Left variable%until it equals to the input In,for filling%the values of Head and Tail.%The enumeration is$,0$,1$,00$,01$,...if Left=In%FinishedRight:=LeftLeft:=$else%Keep enumeratingif Direction=New%default valif Head(Left)=$%$->0$Left:=Cons_0(Left)Head(Cons_0(Left)):=0Tail(Cons_0(Left)):=Leftif Head(Left)=0%e.g.110$->111$Left:=Cons_1(Tail(Left))Head(Cons_1(Tail(Left)):=1Tail(Cons_1(Tail(Left)):=Tail(Left)if Head(Left)=1%01$->10$;11$->000$Direction:=BackwardLeft:=Tail(Left)Right:=Cons_0(Right)if Direction=Backwardif Head(Left)=$%add rightmost digitDirection:=ForwardAddDigit:=Trueif Head(Left)=0%change to1Left:=Cons_1(Tail(Left))Direction:=Forwardif Head(Left)=1%keep backwardsLeft:=Tail(Left)Right:=Cons_0(Right)if Direction=Forward%Gather right0sif Head(Right)=$%finished gatheringDirection:=Newif AddDigit=1Left:=Cons_0(Left)Head(Cons_0(Left)):=0Tail(Cons_0(Left)):=Left7AddDigit=0elseLeft:=Cons_0(Left)Right:=Tail(Right)Head(Cons_0(Left)):=0Tail(Cons_0(Left)):=Left4.2Counter Machines.Counter machines(CM)can be described by the following effective model E:The domain is the natural numbers N.The base structure consists of a nullary function Zero and a unary function Succ,interpreted as the reg-ular successor over N.The almost-constant structure has the vocabulary (name/arity):Out/0,CurrentLine/0,P red/1,Next/1,Reg0,...,Reg n/0, and Line1,...,Line k/0.Its initial data are T rue=1,Line i=i,and all other locations are0.The same structure applies to all machines,except for the number of registers(Reg i)and the number of lines(Line i).For every counter machine m∈CM,define an effective procedure m ∈E with the followingflat program:%Initialization:fill the values of the%predecessor function up to the value%of the inputif CurrentLine=Zeroif Next=Succ(In)CurrentLine:=Line_1elsePred(Succ(Next)):=NextNext:=Succ(Next)%Simulate the counter-machine program.%The values of a,b,c and d are as in%the CM-program lines.if CurrentLine=Line_1Reg_a:=Succ(Reg_a)%or Pred(Reg_a)Pred(Succ(Reg_a)):=Reg_aif Reg_b=ZeroCurrentLine:=celseCurrentLine:=dif CurrentLine=Line_2...%Always:Out:=Reg_085DiscussionIn[3],Gurevich proved that any algorithm satisfying his postulates can be represented by an Abstract State Machine.But an ASM is designed to be “abstract”,so is defined on top of an arbitrary structure that may contain non-effective functions.Hence,it may compute non-effective functions.We have adopted Gurevich’s postulates,but added an additional postulate(Axiom3.3) for effectivity:an algorithm’s initial state may contain onlyfinite data in addition to the domain representation.Different runs of the same procedure share the same initial data,except for the input;different procedures of the same model share a base structure.Here,we showed that Turing machines and counter machines are effective models.In[1],we prove theflip side,namely that Turing machines can sim-ulate all effective models.To cover hypercomputational models,one would need to relax the effectivity axiom or the bounded exploration requirement. References[1]Udi Boker and Nachum Dershowitz,A formalization of the Church-TuringThesis,in preparation.[2]N.G.Fruja and R.F.St¨a rk.The hidden computation steps of Turbo AbstractState Machines.In E.B¨o rger,A.Gargantini,and E.Riccobene,editors,Abstract State Machines—Advances in Theory and Applications,10th International Workshop,ASM2003,Taormina,Italy,pages244–262.Springer-Verlag,Lecture Notes in Computer Science2589,2003.[3]Yuri Gurevich.Sequential abstract state machines capture sequential algorithms.ACM Transactions on Computational Logic,1:77–111,2000.9。
Lecture Notes in Computer Science 1 Mining the Web for Synonyms PMI-IR versus LSA on TOEFL
Lecture Notes in Computer Science 1 Mining the Web for Synonyms:PMI-IR versus LSA on TOEFLPeter D. TurneyInstitute for Information Technology, National Research Council of Canada,M-50 Montreal Road, Ottawa, Ontario, Canada, K1A 0R6peter.turney@nrc.caAbstract. This paper presents a simple unsupervised learning algorithm for rec-ognizing synonyms, based on statistical data acquired by querying a Web searchengine. The algorithm, called PMI-IR, uses Pointwise Mutual Information(PMI) and Information Retrieval (IR) to measure the similarity of pairs ofwords. PMI-IR is empirically evaluated using 80 synonym test questions fromthe Test of English as a Foreign Language (TOEFL) and 50 synonym test ques-tions from a collection of tests for students of English as a Second Language(ESL). On both tests, the algorithm obtains a score of 74%. PMI-IR is con-trasted with Latent Semantic Analysis (LSA), which achieves a score of 64% onthe same 80 TOEFL questions. The paper discusses potential applications of thenew unsupervised learning algorithm and some implications of the results forLSA and LSI (Latent Semantic Indexing).1IntroductionThis paper introduces a simple unsupervised learning algorithm for recognizing syno-nyms. The task of recognizing synonyms is, given a problem word and a set of alter-native words, choose the member from the set of alternative words that is most similar in meaning to the problem word. The unsupervised learning algorithm performs this task by issuing queries to a search engine and analyzing the replies to the queries. The algorithm, called PMI-IR, uses Pointwise Mutual Information (PMI) [1, 2] to analyze statistical data collected by Information Retrieval (IR). The quality of the algorithm’s performance depends on the size of the document collection that is indexed by the search engine and the expressive power of the search engine’s query language. The results presented here are based on queries to the AltaVista search engine [3].Recognizing synonyms is often used as a test to measure a (human) student’s mas-tery of a language. I evaluate the performance of PMI-IR using 80 synonym test ques-tions from the Test of English as a Foreign Language (TOEFL) [4] and 50 synonym test questions from a collection of tests for students of English as a Second Language (ESL) [5]. PMI-IR obtains a score of 73.75% on the 80 TOEFL questions (59/80) and 74% on the 50 ESL questions (37/50). By comparison, the average score on the 80 TOEFL questions, for a large sample of applicants to US colleges from non-EnglishLecture Notes in Computer Science 2 speaking countries, was 64.5% (51.6/80) [6]. Landauer and Dumais [6] note that, “…we have been told that the average score is adequate for admission to many universi-ties.”Latent Semantic Analysis (LSA) is another unsupervised learning algorithm that has been applied to the task of recognizing synonyms. LSA achieves a score of 64.4% (51.5/80) on the 80 TOEFL questions [6]. Landauer and Dumais [6] write, regarding this score for LSA, “We know of no other fully automatic application of a knowledge acquisition and representation model, one that does not depend on knowledge being entered by a human but only on its acquisition from the kinds of experience on which a human relies, that has been capable of performing well on a full scale test used for adults.” It is interesting that PMI-IR, which is conceptually simpler than LSA, scores almost 10% higher on the TOEFL questions.LSA is a statistical algorithm based on Singular Value Decomposition (SVD). A variation on this algorithm has been applied to information retrieval, where it is known as Latent Semantic Indexing (LSI) [7]. The performance of LSA on the TOEFL test has been widely cited as evidence for the value of LSA and (by relation) LSI. In this paper, I discuss the implications of the new unsupervised learning algorithm and the synonym test results for LSA and LSI.In the next section, I describe the PMI-IR algorithm. I then discuss related work on synonym recognition in Section 3. I briefly explain LSA in the following section. The experiments with the TOEFL questions and the ESL questions are presented in Sec-tions 5 and 6, respectively. In Section 7, I discuss the interpretation of the results and their significance for LSA and LSI. The next section discusses potential applications of PMI-IR and the final section gives the conclusions.2PMI-IRConsider the following synonym test question, one of the 80 TOEFL questions. Given the problem word levied and the four alternative words imposed, believed, requested, correlated, which of the alternatives is most similar in meaning to the problem word[8]? Let problem represent the problem word and {choice1, choice2, …, choicen} repre-sent the alternatives. The PMI-IR algorithm assigns a score to each choice, score(choicei), and selects the choice that maximizes the score.The PMI-IR algorithm, like LSA, is based on co-occurrence [9]. The core idea is that “a word is characterized by the company it keeps” [10]. There are many different measures of the degree to which two words co-occur [9]. PMI-IR uses Pointwise Mu-tual Information (PMI) [1, 2], as follows:score(choicei ) = log2(p(problem & choicei) / (p(problem)p(choicei)))(1)Here, p(problem & choicei ) is the probability that problem and choiceico-occur. Ifproblem and choiceiare statistically independent, then the probability that they co-occur is given by the product p(problem)p(choicei). If they are not independent, and they have a tendency to co-occur, then p(problem & choicei) will be greater thanLecture Notes in Computer Science 3p(problem)p(choicei ). Therefore the ratio between p(problem & choicei) andp(problem)p(choicei) is a measure of the degree of statistical dependence between problem and choicei. The log of this ratio is the amount of information that we acquire about the presence of problem when we observe choicei. Since the equation is sym-metrical, it is also the amount of information that we acquire about the presence of choiceiwhen we observe problem, which explains the term mutual information.1Since we are looking for the maximum score, we can drop log2 (because it ismonotonically increasing) and p(problem) (because it has the same value for all choices, for a given problem word). Thus (1) simplifies to:score(choicei ) = p(problem & choicei) / p(choicei)(2)In other words, each choice is simply scored by the conditional probability of the problem word, given the choice word, p(problem | choicei).PMI-IR uses Information Retrieval (IR) to calculate the probabilities in (2). In this paper, I evaluate four different versions of PMI-IR, using four different kinds of que-ries. The following description of these four different methods for calculating (2) uses the AltaVista Advanced Search query syntax [11]. Let hits(query) be the number of hits (the number of documents retrieved) when the query query is given to AltaVista. The four scores are presented in order of increasing sophistication. They can be seen as increasingly refined interpretations of what it means for two words to co-occur, or increasingly refined interpretations of equation (2).Score 1: In the simplest case, we say that two words co-occur when they appear in the same document:score1(choicei) = hits(problem AND choicei) / hits(choicei)(3)We ask AltaVista how many documents contain both problem and choicei, and then we ask how many documents contain choiceialone. The ratio of these two numbers is the score for choicei.Score 2: Instead of asking how many documents contain both problem and choicei, we can ask how many documents contain the two words close together:score2(choicei) = hits(problem NEAR choicei) / hits(choicei)(4)The AltaVista NEAR operator constrains the search to documents that contain prob-lem and choiceiwithin ten words of one another, in either order.Score 3: The first two scores tend to score antonyms as highly as synonyms. For ex-ample, big and small may get the same score as big and large. The following score tends to reduce this effect, resulting in lower scores for antonyms:1 For an explanation of the term pointwise mutual information, see [9].Lecture Notes in Computer Science 4score 3(choice i ) =hits((problem NEAR choice i ) AND NOT ((problem OR choice i ) NEAR "not")) hits(choice i AND NOT (choice i NEAR "not"))(5)Score 4: The fourth score takes context into account. There is no context for the TOEFL questions, but the ESL questions involve context. For example [5], “Every year in the early spring farmers [tap] maple syrup from their trees (drain; boil; knock;rap).” The problem word tap , out of context, might seem to best match the choice words knock or rap , but the context maple syrup makes drain a better match for tap . In general, in addition to the problem word problem and the alternatives {choice 1,choice 2, …, choice n }, we may have context words {context 1, context 2, …, context m }.The following score includes a context word:score 4(choice i ) =hits((problem NEAR choice i ) AND context AND NOT ((problem OR choice i ) NEAR "not"))hits(choice i AND context AND NOT (choice i NEAR "not"))(6)This equation easily generalizes to multiple context words, using AND, but each addi-tional context word narrows the sample size, which might make the score more sensi-tive to noise (and could also reduce the sample size to zero). To address this issue, I chose only one context word from each ESL question. For a given ESL question, I automatically selected the context word by first eliminating the problem word (tap ),the alternatives (drain, boil, knock, rap ), and stop words (in, the, from, their ). The remaining words (every, year, early, spring, farmers, maple, syrup, trees ) were con-text words. I then used p(problem | context i ), as calculated by score 3(context i ), toevaluate each context word. In this example, syrup had the highest score (maple was second highest; that is, maple and syrup have the highest semantic similarity to tap ,according to score 3), so syrup was selected as the context word context for calculatingscore 4(choice i ).3Related WorkThere are several well-known lexical database systems that include synonym informa-tion, such as WordNet [12], BRICO [13], and EuroWordNet [14]. These systems were constructed by hand, without machine learning, which ensures a certain level of qual-ity, at the cost of a substantial amount of human labour. A major limitation of such hand-generated lexicons is the relatively poor coverage of technical and scientific terms. For example, I am interested in applying synonym recognition algorithms to theLecture Notes in Computer Science 5 automatic extraction of keywords from documents [15]. In a large collection of scien-tific and technical journals, I found that only about 70% of the authors’ keywords were in WordNet. (On the other hand, 100% were indexed by AltaVista.) This is a strong motivation for automating aspects of the construction of lexical databases. Another motivation is that the labour involved must be repeated for each new language and must be repeated regularly as new terms are added to a language.Statistical approaches to synonym recognition are based on co-occurrence [9]. Manning and Schütze distinguish between co-occurrence (or association) and colloca-tion: collocation refers to “grammatically bound elements that occur in a particular order”, but co-occurrence and association refer to “the more general phenomenon of words that are likely to be used in the same context” [9]. Order does not matter for synonyms, so we say that they co-occur, rather than saying that they are collocated. Pointwise Mutual Information (PMI) has primarily been applied to analysis of collo-cation, but there have been some applications to co-occurrence analysis [1, 2]. I be-lieve that the novelty in PMI-IR is mainly the observation that PMI can exploit IR. Instead of analyzing a document collection from scratch, specifically for co-occurrence information, we can take advantage of the huge document collections that have been indexed by modern Web search engines.Various measures of semantic similarity between word pairs have been proposed, some using statistical (unsupervised learning from text) techniques [16, 17, 18], some using lexical databases (hand-built) [19, 20], and some hybrid approaches, combining statistics and lexical information [21, 22]. Statistical techniques typically suffer from the sparse data problem: they perform poorly when the words are relatively rare, due to the scarcity of data. Hybrid approaches attempt to address this problem by supple-menting sparse data with information from a lexical database [21, 22]. PMI-IR ad-dresses the sparse data problem by using a huge data source: the Web. As far as I know, no previous work in the statistical approach to semantic similarity has been able to exploit such a large body of text.Another popular statistical approach to measuring semantic similarity is Latent Se-mantic Analysis (LSA) [6, 7, 8]. I will discuss this approach in the next section.The work described in this paper is also related to the literature on data mining and text mining, in that it presents a method for extracting interesting relational informa-tion from a very large database (AltaVista). The most closely related work is the use of interest to discover interesting associations in large databases [23]. The interest of an association A & B is defined as p(A & B) / (p(A)p(B)). This is clearly equivalent to PMI without the log function (see equation (1) above). As far as I know, interest has been applied to data mining, but not to text mining.4Latent Semantic AnalysisLSA uses the Singular Value Decomposition (SVD) to analyze the statistical relation-ships among words in a collection of text [6, 7, 8]. The first step is to use the text to construct a matrix X, in which the row vectors represent words and the column vectorsLecture Notes in Computer Science 6 represent chunks of text (e.g., sentences, paragraphs, documents). Each cell represents the weight of the corresponding word in the corresponding chunk of text. The weight is typically the TF.IDF score (Term Frequency times Inverse Document Frequency) for the word in the chunk. (TF.IDF is a standard tool in Information Retrieval.) The next step is to apply SVD to X, to decompose X into a product of three matrices ULA T, where U and A are in column orthonormal form (i.e., the columns are orthogo-nal and have unit length) and L is a diagonal matrix of singular values (hence SVD). If X is of rank r, then L is also of rank r. Let Lk, where k < r, be the matrix produced by removing from L the r - k columns and rows with the smallest singular values, andlet Uk and Akbe the matrices produced by removing the corresponding columns fromU and A. The matrix Uk LkAkT is the matrix of rank k that best approximates the origi-nal matrix X, in the sense that it minimizes the sum of the squares of the approxima-tion errors. We may think of this matrix Uk LkAkT as a “smoothed” or “compressed”version of the original matrix X. SVD may be viewed as a form of principal compo-nents analysis. LSA works by measuring the similarity of words using this compressedmatrix, instead of the original matrix. The similarity of two words is measured by the cosine of the angle between their corresponding compressed row vectors.When they applied LSA to the TOEFL questions, Landauer and Dumais used an encyclopedia as the text source, to build a matrix X with 61,000 rows (words) and30,473 columns (chunks of text; each chunk was one article from the encyclopedia) [6, 8]. They used SVD to generate a reduced matrix of rank 300. When they measured the similarity of the words (row vectors) in the original matrix X, only 36.8% of theTOEFL questions were answered correctly (15.8% when corrected for guessing, using a penalty of 1/3 for each incorrect answer), but using the reduced matrix of rank 300 improves the performance to 64.4% (52.5% corrected for guessing). They claim that the score of 36.8%, using the original matrix, “… is similar to what would be obtained by a mutual information analysis…” (see footnote 5 in [6]).5TOEFL ExperimentsRecall the sample TOEFL question: Given the problem word levied and the four alter-native words imposed, believed, requested, correlated, which of the alternatives ismost similar in meaning to the problem word [8]? Table 1 shows in detail how score3 is calculated for this example. In this case, PMI-IR selects imposed as the answer.Table 2 shows the scores calculated by LSA for the same example [8]. Note that LSA and AltaVista are using quite different document collections for their calcula-tions. AltaVista indexes 350 million web pages [24] (but only a fraction of them are in English). To apply LSA to the TOEFL questions, an encyclopedia was used to create a matrix of 61,000 words by 30,473 articles [8]. However, it is interesting that the two techniques produce identical rankings for this example.Table 3 shows the results for PMI-IR, for the first three scores, on the 80 TOEFL questions. (The fourth score is not applicable, because there is no context for the questions.) The results for LSA and humans are also presented, for comparison.Lecture Notes in Computer Science 7Table 1. Details of the calculation of score3for a sample TOEFL question.Query Hits imposed AND NOT (imposed NEAR "not")1,147,535 believed AND NOT (believed NEAR "not")2,246,982 requested AND NOT (requested NEAR "not")7,457,552 correlated AND NOT (correlated NEAR "not")296,631 (levied NEAR imposed) AND NOT ((levied OR imposed) NEAR "not")2,299 (levied NEAR believed) AND NOT ((levied OR believed) NEAR "not")80 (levied NEAR requested) AND NOT ((levied OR requested) NEAR "not")216 (levied NEAR correlated) AND NOT ((levied OR correlated) NEAR "not")3Choice Score3 p(levied | imposed)2,299 / 1,147,5350.0020034 p(levied | believed)80 / 2,246,9820.0000356 p(levied | requested)216 / 7,457,5520.0000290 p(levied | correlated) 3 / 296,6310.0000101Table 2. LSA scores for a sample TOEFL question.Choice LSA Scoreimposed0.70believed0.09requested0.05correlated-0.03Table 3. Results of the TOEFL experiments, including LSA results from [6]. Interpretation ofp(problem | choicei )Description ofInterpretationNumberof CorrectTest AnswersPercentage ofCorrect Answersscore1co-occurrenceusing ANDoperator50/8062.5%score2co-occurrenceusing NEAR58/8072.5%score3co-occurrenceusing NEARand NOT59/8073.75%Latent Semantic Analysis51.5/8064.4% Average Non-English US CollegeApplicant51.6/8064.5%Lecture Notes in Computer Science 8 6ESL ExperimentsTo validate the performance of PMI-IR on the TOEFL questions, I obtained another set of 50 synonym test questions [5]. Table 4 shows the results of PMI-IR using all four of the different interpretations of p(problem | choicei).Table 4. Results of the ESL experiments.Interpretation ofp(problem | choicei )Description ofInterpretationNumberof CorrectTest AnswersPercentage ofCorrect Answersscore1co-occurrenceusing ANDoperator24/5048%score2co-occurrenceusing NEAR31/5062%score3co-occurrenceusing NEARand NOT33/5066%score4co-occurrenceusing NEAR,NOT, and context37/5074%7Discussion of ResultsThe results with the TOEFL questions show that PMI-IR (in particular, score3) canscore almost 10% higher than LSA. The results with the ESL questions support the view that this performance is not a chance occurrence. However, the interpretation of the results is difficult, due to two factors: (1) PMI-IR is using a much larger datasource than LSA. (2) PMI-IR (in the case of all of the scores except for score1) isusing a much smaller chunk size than LSA.PMI-IR was implemented as a simple, short Perl program. One TOEFL question requires eight queries to AltaVista (Table 1).2 Each query takes about two seconds, for a total of about sixteen seconds per TOEFL question. Almost all of the time is spent on network traffic between the computer that hosts PMI-IR and the computer(s) that host(s) AltaVista. If PMI-IR were multi-threaded, the eight queries could be is-sued simultaneously, cutting the total time to about two seconds per TOEFL question. If PMI-IR and AltaVista were hosted on the same computer, the time per TOEFL question would likely be a small fraction of a second. Clearly, the hard work here is done by AltaVista, not by the Perl program.2 For the ESL questions, score4requires extra queries to select the context word.Lecture Notes in Computer Science 9 The majority of the time required for LSA is the time spent on the SVD. To com-press the 61,000 by 30,473 matrix used for the TOEFL questions to a matrix of rank 300 required about three hours of computation on a Unix workstation [6]. A fast SVD algorithm can find a rank k approximation to an m by n matrix X in time O(mk2) [25]. Recall that m is the number of words and n is the number of chunks of text. If we suppose that there are about one million English words, then to go from m≈ 50,000 to m≈ 1,000,000 is an increase by a factor of 20, so it seems possible for SVD to be applied to the same corpus as AltaVista, 350 million web pages [24]. For future work, it would be interesting to see how LSA performs with such a large collection of text.Several authors have observed that PMI is especially sensitive to the sparse data problem [9]. Landauer and Dumais claim that mutual information analysis would obtain a score of about 37% on the TOEFL questions, given the same source text and chunk size as they used for LSA (footnote 5 in [6]). Although it appears that they have not tested this conjecture, it seems plausible to me. It seems likely that PMI-IR achieves high performance by “brute force”, through the sheer size of the corpus of text that is indexed by AltaVista. It would be interesting to test this hypothesis. Al-though it might be a challenge to scale LSA up to this volume of text, PMI can easily be scaled down to the encyclopedia text that is used by Landauer and Dumais [6]. This is another possibility for future work. Perhaps the strength of LSA is that it can achieve relatively good performance with relatively little text. This is what we would expect from the “smoothing” or “compression” produced by SVD. However, if you have access to huge volumes of data, there is much less need for smoothing.It is interesting that the TOEFL performance for score1 (62.5%) is approximatelythe same as the performance for LSA (64.4%) (Table 3). Much of the difference in performance between LSA and PMI-IR comes from using the NEAR operator instead of the AND operator. This suggests that perhaps much of the difference between LSA and PMI-IR is due to the smaller chunk size of PMI-IR (for the scores other thanscore1). To test this hypothesis, the LSA experiment with TOEFL could be repeatedusing the same source text (an encyclopedia), but a smaller chunk size. This is another possibility for future work.Latent Semantic Indexing (LSI) applies LSA to Information Retrieval. The hope is that LSI can improve the performance of IR by, in essence, automatically expanding a query with synonyms [7]. Then a search for (say) cars may be able to return a docu-ment that contains automobiles, but not cars. Although there have been some positive results using LSI for IR [8], the results from TREC2 and TREC3 (Text Retrieval Con-ferences 2 and 3) did not show an advantage to LSI over other leading IR techniques [26]. It has been conjectured that the TREC queries are unusually long and detailed, so there is little room for improvement by LSI [8]. The results reported here for PMI-IR suggest an alternative hypothesis. Most of the TREC systems use a technique called query expansion [27]. This technique involves searching with the original query, ex-tracting terms from the top retrieved documents, adding these terms to the original query, and then repeating the search with the new, expanded query. I hypothesize that this query expansion achieves essentially the same effect as LSI, so there is no appar-ent advantage to LSI when it is compared to an IR system that uses query expansion.Lecture Notes in Computer Science 10 If (say) cars and automobiles have a high semantic similarity, then we can expect p(automobiles | cars) to be relatively high (see equation (2)). Thus, the query cars is likely to retrieve a document containing the word automobiles. This means that there is a good chance that query expansion will expand the query cars to a new query that contains automobiles. Testing this hypothesis is another area for future work. The hypothesis implies that LSI will tend to perform better than an IR system without query expansion, but there will be no significant difference between an IR system with LSI and an IR system with query expansion (assuming all other factors are equal).8ApplicationsA limitation of PMI-IR is that the network access time for querying a large Web search engine may be prohibitive for certain applications, for those of us who do not have very high-speed, high-priority access to such a search engine. However, it is possible that PMI-IR may achieve good results with a significantly smaller document collection. One possibility is a hybrid system, which uses a small, local search engine for high-frequency words, but resorts to a large, distant search engine for rare words.PMI-IR may be suitable as a tool to aid in the construction of lexical databases. It might also be useful for improving IR systems. For example, an IR system with queryexpansion might use score4 to screen candidate terms for expanding a query. The can-didates would be extracted from the top retrieved documents, as with current query expansion techniques. However, current query expansion techniques may suggest sub-optimal expansion terms, because the top retrieved documents constitute a relatively small, noisy sample. Thus there could be some benefit to validating the suggested expansions using PMI-IR, which would draw on larger sample sizes.I am particularly interested in applying PMI-IR to automatic keyword extraction[15]. One of the most helpful clues that a word (or phrase) is a keyword in a given document is the frequency of the word. However, authors often use synonyms, in order to avoid boring the reader with repetition. This is courteous for human readers, but it complicates automatic keyword extraction. I am hoping that PMI-IR will help me to cluster synonyms together, so that I can aggregate their frequency counts, re-sulting in better keyword extraction.9ConclusionsThis paper has introduced a simple unsupervised learning algorithm for recognizing synonyms. The algorithm uses a well-known measure of semantic similarity (PMI). The new contribution is the observation that PMI can exploit the huge document col-lections that are indexed by modern Web search engines. The algorithm is evaluated using the Test of English as a Foreign Language. The algorithm is compared with Latent Semantic Analysis, which has also been evaluated using TOEFL. The compari-。
lncs 中文参考文献格式
lncs 中文参考文献格式
在Lecture Notes in Computer Science (LNCS)中，中文参考文献的格式如下：
1. 书籍的格式：
[序号] 作者姓氏，作者名字. 书名. 版本号（可选）. 出版地点：出
版社，出版年份.
示例：
[1] 李明，王小红. 计算机科学导论. 北京：清华大学出版社，2021.
2. 期刊文章的格式：
[序号] 作者姓氏，作者名字. "文章标题." 期刊名，卷号(期号)，页码，出版年份.
示例：
[1] 张华，李雷. "基于人工智能的图像识别技术." 计算机科学与技
术学报，28(2)，377-391，2020.
3. 会议论文的格式：
[序号] 作者姓氏，作者名字. "论文标题." 会议名，页码，出版地点，出版年份.
示例：
[1] 王志刚，张华. "面向物联网的智能家居系统设计." 中国计算机
大会，203-210，北京，2019.
4. 网上资源（例如学位论文、技术报告）的格式：
[序号] 作者姓氏，作者名字. "文章标题." 在线资源类型，可获取的
网站或数据库，年份.
示例：
[1] 赵明，李丽. "基于深度学习的图像分类技术." 学位论文，清华
大学学位论文库，2020.
请注意，根据具体情况，可能会有略微的变化。
为了符合LNCS
的格式要求，请确保正确引用文献的相关信息，并遵守相关规定。
Lecture Notes in Computer Science, Volume 936)
{ Kahn semantics applies to static networks, whereas the pi{calculus can
encode dynamic networks. The solution to this problem is motivated by tools used in the study of computation in Linear Logic 8], in particular the Geometry of Interaction program 9, 1]. For the purposes of this paper, the key relevant idea is to mimic dynamic networks (of the lambda calculus) by ow of structured tokens in a static network. { The second complication that arises is the non-determinism in pi{calculus processes. The approach we adopt here is to use a variant of the generalization of the determinate data- ow semantics to a semantics for indeterminate data ow networks 22, 2].
By changing the structure of the tokens, the resulting semantics applies uniformly to the Calculus of Communicating Systems 19] and the pi{calculus. This change of structure of tokens is suggested by the nature of computation in the pi{calculus. The de nitions of the process combinators essentially stand unchanged. In particular, the de nition of parallel composition is unchanged. Thus our treatment of interaction is \generic", an essential criterion for any good description of parallel composition. Furthermore, for all of these calculi, our semantics distinguishes the processes a k b and ab + ba. More generally, for CCS, we show that our semantics induces the same process equivalence as a pomset-based semannger-Verlag Berlin Heidelberg 1995. This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting, reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer-Verlag. Violations are liable for prosecution under the German Copyright Law.
浙江大学教师岗位聘任
浙江大学教师岗位聘任申请表*名：***学院（系、平台等）：理学院现专业技术职务：教授现专业技术岗位等级：2级教师岗位类别：9级岗申报级别：B2所属学部：理学部年月日附件一：浙江大学2010年申报教师岗位人员主要业绩（2006.1—2009.12）学院（系、平台等）：理学院姓名：彭群生性别：男出生年月：1947.5所在二级学科：应用数学最后学历及毕业时间：博士，1983年9月现专业技术职务及任职时间：教授，1988年博导时间：1993年现专业技术岗位等级：2级岗位聘任级别：9级兼任党政职务：曾担任民进浙大委员会主委（2003－2008），浙江省政协委员（2003－2007）。
主要学术兼职：担任中国计算机学会CAD与图形学专委会主任，《The Visual Computer》、《J. Computer Science & Technology》、《中国科学－信息科学》、《计算机学报》、《软件学报》、《计算机辅助设计与图形学学报》、《J of Zhejiang University- Science 》等7种国内外期刊编委，浙江大学CAD&CG国家重点实验室学术委员会副主任、北京航空航天大学虚拟现实系统国家重点实验室学术委员、复旦大学智能信息处理重点实验室学术委员。
现教师岗位类别：教学科研并重岗申报岗位聘任级别：B2（一）教学工作：1、主讲课程名称、课程类别（如本科生通识课程、大类课程、专业课程，研究生公共课程等）、授课对象、学生数、学时数、教学年度、考核结果计算机图形学，研究生学位课，研究生， 100人，72学时/年，每年秋－冬学期2、承担大学生体育文化活动指导等（列出教学年度、主要内容、授课/指导对象、人数（次）、学时数）3、指导本科生毕业论文（设计）22人(列出本科生专业、年级、姓名)数学2002级：张鑫，徐小东，金志栋；计算机2002级：杨颖振，谢立广，杨志亮数学2003级陈佳舟，程军，于洋；计算机2003级：陈曦，任智敏，徐小华，丁子昂，骆鹏程数学2004级石峰；计算机2004级：李路莹，黄若冠，刘华航数学2005级林乃养；计算机2005级：莫铭臻，段鑫，郗加河4、本科生导师工作情况，指导22名(列出所学专业、年级、姓名、考核等级)数学2003级陈佳舟，程军，于洋；计算机2003级：陈曦，任智敏，徐小华，丁子昂，骆鹏程数学2004级石峰；计算机2004级：李路莹，黄若冠，刘华航，数学2005级林乃养；计算机2005级：莫铭臻，段鑫，郗加河数学2006级鲁佳，金烁；计算机2006级：方婧，陈曦，王冉，陈翔5、指导博士生29 名、硕士生36名(列出研究生(不含研究生课程进修班学员)所学专业、年级、姓名)博士生:应用数学2002：吴向阳，柴登峰，缪永伟，王长波，应用数学2003：郭延文，肖春霞，江照意，刘世光，张龙，管宇，张涛；计算机科学与技术2003: 宋成芳，胡敏应用数学2004：刘春晓，汪莉计算机科学与技术2004：张繁，应用数学2005：钟凡，延诃，张元慧，潘斌，王锐，应用数学2006：张鑫，张艺江，韩玮，刘艳丽，赵勇，计算机科学与技术2006：范涵奇应用数学2007 陈佳舟；应用数学2008：张艺江，计算机科学与技术2009：张祯硕士生：应用数学2003：武凤霞；应用数学2004：李辉，潘梁，刘海芹，龙珑计算机科学与技术2004：范亦楠，马瑞金，龚怿，应逸亭，曾运，应用数学2005：刘舒，徐小东，王庆伟，涂晓兰；计算机科学与技术2005：虞宏毅，谈奇峰，陈飞飞应用数学2006：金志栋，姚芸，林成春；计算机科学与技术2006：罗功辉，杨志亮，杨颖振，应用数学2007：程军，于洋：计算机科学与技术2007：何戬，陈洪文，丁子昂，姚建应用数学2008：王智广，邢冠宇；计算机科学与技术2008：杨中雷应用数学2009：林乃养，崔晓燕；计算机科学与技术2009：王帅、袁霞（三）主持或主参（前三位）的科研项目项目名称、项目来源、项目编号、起止时间、经费总额本人排名/总人数1. 虚拟环境的统一信息表达理论与高效构建方法，国家973项目，2002CB312101，2003.1～2007.12，440万，1/10，2．蛋白质结构的分子场建模、表达与分析，国家自然科学基金（重点项目），60533050，2006.1～2009.12，200万，1/10，3．增强虚拟型混合环境的呈现，国家973项目，2009CB320802，2009.1～2013.12，495万，2/10， 4．场景表意式绘制方法研究，国家自然科学基金（面上项目）2010.1～2012.12， 32万，1/ 8（四）论文著作1、发表论文共 106 篇（列出论文题目、所载刊物、影响因子、他引次数、发表年月、本人排名/总人数），如Top期刊请注明。
Springer_CS_Proceedings_Author_Guidelines_19APR2011
Author Guidelines for the Preparation of Contributionsto Springer Computer Science ProceedingsAlfred Hofmann 1,*{alfred.hofmann,ralf.gerstner,anna.kramer , Ralf Gerstner 1, Anna Kramer 1, and Frank Holzwarth 21Springer-Verlag, Computer Science Editorial, Heidelberg, Germany}@2 Springer-Verlag, Technical Support, Heidelberg, Germanyfrank.holzwarth@ Abstract. The abstract is a mandatory element that should summarize the con-tents of the paper and should contain at least 70 and at most 150 words. Ab-stract and keywords are freely available in SpringerLink.Keywords: We would like to encourage you to list your keywords here. They should be separated by commas.1 IntroductionYou will find here Springer’s guidelines for the preparation of proceedings papers to be published in one of the following series, in printed and electronic form:• Lecture Notes in Computer Science (LNCS), incl. its subseries Lecture Notes in Artificial Intelligence (LNAI) and Lecture Notes in Bioinformatics (LNBI), and LNCS Transactions; • Lecture Notes in Business Information Processing (LNBIP);• Communications in Computer and Information Science (CCIS);• Lecture Notes of the Institute for Computer Sciences, Social Informatics and Tele-communications Engineering (LNICST);• IFIP Advances in Information and Communication Technology (IFIP AICT), for-merly known as the IFIP Series;• Proceedings in Information and Communications Technology (PICT).Your contribution may be prepared in LaTeX or Microsoft Word. Technical Instruc-tions for working with Springer’s style files and templates are provided in separate documents which can be found in the respective zip packages on our website. * No academic titles or descriptions of academic positions should be included in the addresses. Either this information should be omitted altogether (preferably), or it should be included in a footnote at the end of the first page. Information of this nature, given in the addresses, will be deleted by our typesetters.22Preparation of Your Paper2.1Structuring Your PaperAffiliations. The affiliated institutions are to be listed directly below the names of the authors. Multiple affiliatons should be marked with superscript arabic numbers, and they should each start on a new line as shown in this document. In addition to the name of your affiliation, we would ask you to give the town and the country in which it is situated. If you prefer to include the entire postal address, then please feel free to do so. E-mail addresses should start on a new line and should be grouped per affiliation. Headings.Headings should be capitalized (i.e., nouns, verbs, and all other words except articles, prepositions, and conjunctions should be set with an initial capital) and should, with the exception of the title, be aligned to the left. Only the first two levels of section headings should be numbered, as shown in Table 1. The respective font sizes are also given in Table 1. Kindly refrain from using “0” when numbering your section headings.Table 1. Font sizes of headings. Table captions should always be positioned above the tables. Heading level Example Font size and style Title (centered) Lecture Notes14 point, bold1st-level heading 1 Introduction12 point, bold2nd-level heading 2.1 Printing Area10 point, bold3rd-level heading Run-in Heading in Bold. Text follows10 point, bold4th-level heading Lowest Level Heading. Text follows10 point, italic Words joined by a hyphen are subject to a special rule. If the first word can stand alone, the second word should be capitalized.Here are some examples of headings: “Criteria to Disprove Context-Freeness of Collage Languages”, “On Correcting the Intrusion of Tracing Non-deterministic Pro-grams by Software”, “A User-Friendly and Extendable Data Distribution System”, “Multi-flip Networks: Parallelizing GenSAT”, “Self-determinations of Man”. Lemmas, Propositions, and Theorems.The numbers accorded to lemmas, proposi-tions, and theorems, etc. should appear in consecutive order, starting with Lemma 1. Please do not include section counters in the numbering like “Theorem 1.1”.2.2Page Numbering and Running HeadsThere is no need to include page numbers or running heads; this will be done at our end. If your paper title is too long to serve as a running head, it will be shortened. Your suggestion as to how to shorten it would be most welcome.32.3 Figures and TablesIt is essential that all illustrations are as clear and as legible as possible. Vector graph-ics – instead of rasterized images – should be used for diagrams and schemas when-ever possible. Please check that the lines in line drawings are not interrupted and have a constant width. Grids and details within the figures must be clearly legible and may not be written one on top of the other. Line drawings are to have a resolution of at least 800 dpi (preferably 1200 dpi). The lettering in figures should not use font sizesp o w e r f l u c t u a t i o n (1555 n m ), d B time, msFig. 1. Power distribution of channel at 1555 nm along the link of 383 km (Source: LNCS 5412, p. 323)Fig. 2. Artifacts empowered by Artificial Intelligence (Source: LNCS 5640, p. 115)4smaller than 6 pt (~ 2mm character height). Figures are to be numbered and to have a caption which should always be positioned under the figures, in contrast to the caption belonging to a table, which should always appear above the table.Captions are set in 9-point type. If they are short, they are centered between the margins. Longer captions, covering more than one line, are justified (Fig. 1and Fig. 2 show examples). Captions that do not constitute a full sentence, do not have a period.Text fragments of fewer than four lines should not appear at the tops or bottoms of pages, following a table or figure. In such cases, it is better to set the figures right at the top or right at the bottom of the page. A figure should never be placed in the mid-dle of a paragraph.If screenshots are necessary, please make sure that the essential content is clear to the reader.Remark 1.In the printed volumes, illustrations are generally black and white (half-tones), and only in exceptional cases, and if the author or the conference organization is prepared to cover the extra costs involved, are colored pictures accepted. Colored pictures are welcome in the electronic version free of charge. If you send colored figures that are to be printed in black and white, please make sure that they really are also legible in black and white. Some colors show up very poorly when printed in black and white.2.4FormulasDisplayed equations or formulas are centered and set on a separate line (with an extra line or half line space above and below). Displayed expressions should be numbered for reference. The numbers should be consecutive within the contribution, with num-bers enclosed in parentheses and set on the right margin. Please do not include section counters in the numbering.x + y = z(1) Equations should be punctuated in the same way as ordinary text but with a small space before the end punctuation mark.2.5FootnotesThe superscript numeral used to refer to a footnote appears in the text either directly after the word to be discussed or – in relation to a phrase or a sentence – following the punctuation mark (comma, semicolon, or period).1For remarks pertaining to the title or the authors’ names, in the header of a paper, symbols should be used instead of a number (see first page of this document). Please note that no footnotes may be included in the abstract.1The footnote numeral is set flush left and the text follows with the usual word spacing.5 2.6Program CodeProgram listings or program commands in the text are normally set in typewriter font: program Inflation (Output){Assuming annual inflation rates of 7%, 8%, and10%,... years};const MaxYears = 10;var Year: 0..MaxYears;Factor1, Factor2, Factor3: Real;beginYear := 0;Factor1 := 1.0; Factor2 := 1.0; Factor3 := 1.0;WriteLn('Year 7% 8% 10%'); WriteLn;repeatYear := Year + 1;Factor1 := Factor1 * 1.07;Factor2 := Factor2 * 1.08;Factor3 := Factor3 * 1.10;WriteLn(Year:5,Factor1:7:3,Factor2:7:3,Factor3:7:3)until Year = MaxYearsend.[Example of a computer program from Jensen K., Wirth N.: Pascal User Manual and Report. Springer, New York (1991)]2.7Citations and BibliographyFor citations in the text, we prefer the use of square brackets and consecutive num-bers. Numbers should be grouped where appropriate. We would write [1-5] but [1], [3], [5], etc. The numbers in the bibliography section are without square brackets. Citations using labels or the author/year convention are also acceptable.Please write all references using the latin alphabet. If the title of the book you are referring to is, e.g., in Russian or Chinese, then please write (in Russian) or (in Chi-nese) at the end of the transcript or translation of the title.In order to permit cross referencing within , and even-tually between different publishers and their online databases, Springer standardizes the format of the references. This feature aims to increase the visibility of publications and facilitate academic research. Please base your references on the examples given in the references section of these instructions. References that do not adhere to this style will be reformatted at our end.We would like to draw your attention to the fact that references to LNCS proceed-ings papers are particularly often reformatted due to missing editor names or incom-plete publisher information. This adjustment may result in the final papers as pub-lished by Springer having more pages than the original versions as submitted by the authors. Here is an example:6Reference as formatted in author’s original version:Assemlal, H.E., Tschumperlé, D., Brun, L.: Efficient Computation of PDF-Based Characteris-tics from Diffusion MR Signal. In: MICCAI. Volume 5242. (2008) 70–78Reference after reformatting by Springer:Assemlal, H.E., Tschumperlé, D., Brun, L.: Efficient Computation of PDF-Based Characteris-tics from Diffusion MR Signal. In: Metaxas, D., Axel, L., Fichtinger, G., Székely, G. (eds.) MICCAI 2008, Part II. LNCS, vol. 5242, pp. 70–78. Springer, Heidelberg (2008)One more line is needed for this reference, as a result of Springer’s adjustment.Please make sure that all your sources are correctly listed in the reference section. Do not include references to pieces of work that are not connected with your paper.The references section at the end of this paper shows a sample reference list with entries for journal articles [1], an LNCS chapter [2], a book [3], proceedings without editors [4] and [5], as well as a URL [6]. Please note that Springer proceedings are cited with their publication acronyms and volume numbers.2.8PlagiarismSpringer takes plagiarism seriously. If an author has copied from another author or has used parts of another author’s work, without his or her permission and a refer-ence, then the paper on SpringerLink will be given a “retracted” stamp, and an erra-tum explaining the reasons for the retraction will be included. In addition, the volume editors and the author’s academic supervisors will be informed. Acknowledgements. This should always be a run-in heading and not a section or subsection heading. It should not be assigned a number. The acknowledgements may include reference to grants or supports received in relation to the work presented in the paper.3Additional Information Required from Authors3.1Copyright FormThere are different copyright forms in place for the different Springer Computer Science proceedings book series. A prefilled copyright form is usually available from the conference website. Alternatively, the form may be downloaded from the web-page of the respective series on . Please send your signed copyright form to your conference publication contact, either as a scanned PDF or by fax or by courier. One author may sign on behalf of all of the other authors of a par-ticular paper. In this case, the author signs for and accepts responsibility for releasing this material on behalf of any and all co-authors. Digital signatures are also accepta-ble.73.2Contact Author InformationKindly assure that, when you submit the final version of your paper, you also provide the name and e-mail address of the contact author for your paper. These details are used to compile a list of contact authors for our typesetting partners SPS in India. The contact author must be available to check the paper roughly seven weeks before the start of the conference, or before the book is due to leave the printing office, in the case of post-conference proceedings. He or she will be contacted by SPS from the following e-mail address: typesetting@sps.co.in.3.3Correct Representation of Author NamesAuthors’ names should be written out in full at the tops of the papers. They are short-ened by us to “initials surname” in the running heads and will take the form “sur-name, given name” in the author index. If you or any of your co-authors have more than one family name, it should be made quite clear how your name is to be displayed in the running heads and the author index. Chinese authors should write their given names in front of their surnames at the tops of their papers. If you only have one (main) name, please make sure that this name is written out in full in the running heads, when you check your final PDF.4Typesetting of Your Paper at SpringerPlease make sure that the paper you submit is final and complete, that any copyright issues have been resolved, that the authors listed at the top of the chapter really are the final authors, and that you have not omitted any references. Following publication, it is not possible to alter your paper on SpringerLink. Kindly note that we prefer the use of American English.4.1What Will Be Done with Your PaperIf the instructions have been followed closely, then only very minor alterations will be made to your paper. The production team at SPS checks the format of the paper, and if, for example, vertical spacing has been inserted or removed, making the paper unsightly or difficult to read, then this is remedied. In addition, running-heads, final page numbers and a copyright line are inserted, and the capitalization of the headings is checked and corrected if need be. Finally, the reference section is attuned to our specifications (see also Section 2.7).4.2Proof Reading StageOnce the files have been worked upon, SPS sends a copy of the final PDF of each paper to its contact author. The contact author is asked to check through the final PDF8to make sure that no errors have crept in during the transfer or preparation of the files. This should not be seen as an opportunity to update or copyedit the paper, which is not possible due to time constraints. Only errors introduced during the preparation of the files will be corrected. Particular attention should be paid to the references section.If SPS does not receive a reply from a particular contact author, within the time-frame given (usually 72 hours), then it is presumed that the author has found no errors in the paper. The tight publication schedule of our proceedings series does not allow SPS to send reminders or search for alternative e-mail addresses on the Internet.In some cases, it is the contact volume editor or the publication chair who checks all of the PDFs. In such cases, the authors are not involved in the checking phase.The purpose of the proof is to check for typesetting or conversion errors and the completeness and accuracy of the text, tables and figures. Substantial changes in con-tent, e.g., new results, corrected values, title and authorship, are not possible and can-not be processed.5Online Publication in SpringerLinkAll papers will be published in our digital library . Only subscribers to Springer’s eBook packages or to the electronic book series are able to access the full text PDFs and references of our online publications. Meta-data, ab-stracts and author e-mail addresses are freely available for all users.6Checklist of Items to Be Sent to Volume Editor•The final source files, incl. any non-standard fonts.• A final PDF file corresponding exactly to the final source files.• A copyright form, signed by one author on behalf of all of the authors of the paper. •The name and e-mail address of the contact author who will check the proof of the paper.• A suggestion for an abbreviated running head, if appropriate.•Information about correct representation of authors’ names, where necessary. References1.Smith, T.F., Waterman, M.S.: Identification of Common Molecular Subsequences. J. Mol.Biol. 147, 195–197 (1981)2.May, P., Ehrlich, H.C., Steinke, T.: ZIB Structure Prediction Pipeline: Composing a Com-plex Biological Workflow through Web Services. In: Nagel, W.E., Walter, W.V., Lehner, W. (eds.) Euro-Par 2006. LNCS, vol. 4128, pp. 1148–1158. Springer, Heidelberg (2006) 3.Foster, I., Kesselman, C.: The Grid: Blueprint for a New Computing Infrastructure. Mor-gan Kaufmann, San Francisco (1999)4.Czajkowski, K., Fitzgerald, S., Foster, I., Kesselman, C.: Grid Information Services forDistributed Resource Sharing. In: 10th IEEE International Symposium on High Perfor-mance Distributed Computing, pp. 181–184. IEEE Press, New York (2001)9 5.Foster, I., Kesselman, C., Nick, J., Tuecke, S.: The Physiology of the Grid: an Open GridServices Architecture for Distributed Systems Integration. Technical report, Global Grid Forum (2002)6.National Center for Biotechnology Information, Appendix: Springer Author DiscountAuthors contributing to any of Springer’s computer science proceedings publications are entitled to a 33.3% discount off all Springer products when placing an order through . Please go to /authors, then click on ‘Book Authors’ and then follow the instructions given in the Book Dis-counts section. You are requested to give full details of your Springer publication to obtain a so-called SpringerToken. This token is a number that must be entered when placing an order through our online catalogue, in order to obtain the discount. Contact UsIf you have any further questions regarding the preparation of your paper, then please do not hesitate to get in touch with us.•For all questions related to our LaTeX style files, your contact person is: Mr. Frank Holzwarth, e-mail: frank.holzwarth@.•For overall technical questions concerning the preparation of LNCS/LNAI/LNBI papers, please contact Ms. Anna Kramer, e-mail: lncs@.•For the LNBIP series, please contact Ms. Viktoria Meyer, e-mail: lnbip@.•For the CCIS series, please contact Ms. Leonie Kunz, e-mail:ccis@.•For the LNICST series, please contact Mr. Peter Strasser, e-mail:lnicst@.•For the IFIP AICT series, please contact Ms. Erika Siebert-Cole, e-mail: ifip@.•For the PICT series, please contact Mr. Masayuki Nakamura, email: pict@springer.jp.[April 19, 2011]。
Springer网络版丛书
Springer网络版丛书一、Springer网络版丛书简介德国施普林格出版社(Springer-Verlag)出版的网络版丛书内容覆盖化学、计算机科学、工程学、环境科学、地理学、生命科学、数学、医学、物理和天文学等学科并包含了很多跨学科内容，丛书间的配合性较好。
已开通的Springer网络版丛书有19种近1900卷，2004年计划出版并在网上开通约500卷（另有2种新丛书计划在2004年上网开通11卷）。
Springer网络版丛书一览表二、Springer网络版丛书的使用方法Springer网络版丛书通过SpringerLink平台提供网络服务。
1．登录SpringerLink在任意一台联网计算机上，不需用户名和密码即可登录SpringerLink的清华镜像站或德国站点。
此时，读者身份为团体用户，可以对包括Springer网络版丛书在内的SpringerLink所有电子出版物进行浏览、检索和查阅文章标题和文摘。
对于SpringerLink提供的一些个性化服务，如收藏夹、最新目录通告等服务，读者须在SpringerLink德国站点主页注册为个人用户并以个人用户登录后方可使用。
中国科学院各研究所、文献情报中心、研究生院等均可在本单位IP范围免费查阅Springer网络版丛书全文。
2．浏览Springer网络版丛书登录SpringerLink后，读者可以通过两种方式进行浏览：（1）利用SpringerLink页面上方的导航条点击导航条内的BROWSE进入浏览页面，选择Publications标签，选中Book Series，点击Browse即进入Browse Results页面，可看到Springer网络版丛书题名列表，逐层向下点击即可进行浏览；（2）利用SpringerLink主页上的Browse Publications链接该链接在SpringerLink主页的Browse and Explore栏目下。
Lecture Notes in Computer
A Bibliography of Papers in Lecture Notes in ComputerScience(1999),Part1of2Nelson H.F.BeebeCenter for Scientific ComputingUniversity of UtahDepartment of Mathematics,322INSCC155S1400E RM233Salt Lake City,UT84112-0090USATel:+18015815254FAX:+18015851640,+18015814148E-mail:beebe@(Internet)WWW URL:/~beebe/05January1999Version1.01Title word cross-reference 2[36].4[34].ABR[46,45].Accommodate[29]. Accountable[26].Adaptive[8]. Address[30].Algorithm[52,46]. Annotation[50].Applets[28]. Application[42,52,33].Applications[30,47,41,27]. Approaches[1].Architecture[36]. Arctic[6].Asynchronously[50].ATM[7,24].Audio[38].Aware[27]. Based[53,13,18,35,36,4,16,32,49,37].Behaviour[54].Better[14].Between[21]. Blocking[22].Browsing[47].Buffers[55]. Can[14].CC[48].CCS[32].CCS1MD[5].Challenges[12].Channels[17].ChaosLAN[20].Chaotic[20].Class[46,2].Classifying[37].Client[49,47].Client-Server[49].Close[21].Clustering[3].Coding[39,35,37]. Collaborative[26].Communication[13,5,21].Comparison[35].Complexity[46]. Compliant[43].Compressed[40]. Computation[5].Computing[39]. Concurrent[5].Conferencing[30,32]. Congestion[3,46].Congruent[38]. Conscious[40].Contemporary[12].12Content[37].Continuous[53,47]. Control[3,46,16,45].Cooperative[47,31].CORBA[32]. CORBA-Based[32].CSCW[41].Cut[4]. Cut-Through[4].Data[53,47,35].DA VIC[48,44].DA VIC-Terminal[48].Deadlock[23,22]. Definition[43].Deflection[3].Delay[41]. Delayed[3].Design[14,20,12,49]. Deterministic[8].Deterministic/Adaptive[8].Direct[19]. Disk[52].Distance[28,16].Distance-Based[16].Distributed[52]. Division[21].Documents[54,49].Does[21].Drop[18].DSM[48,14].DSM-CC-Server[48].Dynamic[42]. Education[28].Efficient[9].Embedded[11].Encryption[35]. Environment[44].Environments[29,37]. Evaluation[13,8].Exchange[30]. Exploiting[54].Fabric[6].Factors[41].Flexible[34]. Flow[16,45].Framework[5].Frequency[23].Gap[21].Gigabit[20].Goes[44].Graphs[51].Guide[27].Guru[25]. Hardware[7].Heterogeneous[44].High[1,9,25].High-Performance[1,9]. HiPER[9].HiPER-P[9].Hybrid[8]. Hypermedia[51].Identifying[38].II[10].Image[39,37]. Images[40].Implementation[48,20]. Importance[41].Injection[23]. Integrated[2,45,31].Integration[15]. Interactive[28,36,49].Interconnection[9,22].Internet[50,44,30].Interworking[44]. Invalidate[13].Invalidate-Based[13].Invited[15].IP[24].Irregular[18,17,4]. Java[49].Key[30].Lab[27].LAN[20].Language[43]. Limiting[23].Location[27].Location-Aware[27].Low[46]. Machines[5].Management[53].Market[27].Mechanisms[13].Media[55,50].Memory[13,55,21]. MESH[33].Message[23,22].Mobile[39,27,37].Modeling[22]. MPEG[34,36].MPEG-2[36].MPEG-4[34].Multi[50,2,47,18].Multi-class[2].Multi-client[47].Multi-Drop[18].Multi-media[50].Multi-server[47].Multicasting[18,4]. Multicomputer[9].Multidestination[18,4].Multimedia[52,44,36,49].Multiplexing[34,21].Multiprocessors[13].Net[15].Network[29,40].Network-Conscious[40].Networking[44].Networks[1,14,40,18, 23,9,19,17,4,16,45,22].NT[15]. Object[43,51].Objectionable[37].ODP[43].ODP-Compliant[43].offs[19]. Optical[21].P[9].Packet[25].Pair[55].Partitioning[35].Path[18].Path-Based[18].Perceptually[38]. Performance[1,29,9,19].Perspective[25].Pivotal[51].Platform[31].PNSVS[42].Power[19]. Power/Performance[19].Prefetching[54].Preliminary[8]. Presentation[15].Project[33].REFERENCES3Protocol[28,30,47,45].QoS[43,42,25].QoS-Support[43]. Quality[1].Real[52].Real-Time[52]. Reconstruction[39].Reducing[55]. Reduction[23].Renegotiation[42]. Replicated[50].Representation[51]. Requirements[55].Reservation[24]. Resolution[30].Retrieval[38].RIP[24]. Routed[3].Router[9,8].Routers[12]. Routing[2,20].Scalable[35].Scheduling[52].Scheme[39].Secure[26].Server[48,49,47].ServerNet[10]. Servers[55].Service[50,1,44,46,32]. Shared[50,13].SIMD[5].Simulation[28].Single[55].Smoothed[7].Spatial[45].Speed[25]. Speeds[21].Standards[11].Storage[53,52].Stored[7].STREAMER[7].Structures[38].Study[41].Support[43,7].Switch[6,18]. Switch-Based[18].Switches[4]. Switching[25].System[36].Systems[53,52,14,34,11].Tape[53].Tape-Based[53].Techniques[29].Teleteaching[31]. Temporal[45].Temporal-Spatial[45]. Terminal[48].Tertiary[53].Their[41]. Time[52,21].Time-Division[21].TINA[33].Topology[17].Torus[3]. Trade[19].Trade-offs[19]. Transmission[7].Tree[4].Tree-Based[4].U[15].U-Net[15].Use[17].User[54]. Using[18,20,4].Varying[29].VBR[55].Video[35,7]. Videoconferencing[42].Virtual[29,17].Visual[29].Websites[37].Whiteboard[26].Win-dows[15].Windows/NT[15].Wire-less[40,37].Workspaces[50].Work-stations[17].Wormhole[3,23,16]. Wormhole-Routed[3].Worms[18,4]. WWW[54,37].ReferencesChien:1999:AQS [1]A.A.Chien.Approaches to quality of ser-vice in high-performance networks.Lec-ture Notes in Computer Science,1417:1–??,1999.CODEN LNCSD9.ISSN0302-9743.Fulgham:1999:IMR [2]M.L.Fulgham and L.Snyder.Inte-grated multi-class routing.Lecture Notesin Computer Science,1417:21–??,1999.CODEN LNCSD9.ISSN0302-9743.Hyatt:1999:CCW [3]C.Hyatt and D.P.Agrawal.Congestioncontrol in the wormhole-routed torus withclustering and delayed deflection.LectureNotes in Computer Science,1417:33–??,1999.CODEN LNCSD9.ISSN0302-9743.Sivaram:1999:MIN [4]R.Sivaram, D.K.Panda,and C. B.Stunkel.Multicasting in irregular net-works with cut-through switches usingtree-based multidestination worms.Lec-ture Notes in Computer Science,1417:39–??,1999.CODEN LNCSD9.ISSN0302-9743.Garg:1999:CCC[5]V.Garg and S1MD:A concurrent communication and com-REFERENCES4putation framework for SIMD machines.Lecture Notes in Computer Science,1417:55–??,1999.CODEN LNCSD9.ISSN0302-9743.Boughton:1999:ASF [6]G.A.Boughton.Arctic switch fabric.Lecture Notes in Computer Science,1417:65–??,1999.CODEN LNCSD9.ISSN0302-9743.Moon:1999:SHS [7]S.-W.Moon,P.Pillai,and K.G.Shin.STREAMER:Hardware supportfor smoothed transmission of stored videoover ATM.Lecture Notes in Com-puter Science,1417:75–??,1999.CODENLNCSD9.ISSN0302-9743.Miller:1999:PEH [8]ler and W.A.Najjar.Prelimi-nary evaluation of a hybrid deterministic/adaptive router.Lecture Notes in Com-puter Science,1417:89–??,1999.CODENLNCSD9.ISSN0302-9743.May:1999:HEH [9]P.May,S.M.Chai,and D.S.Wills.HiPER-P:An efficient,high-performancerouter for multicomputer interconnectionnetworks.Lecture Notes in ComputerScience,1417:103–??,1999.CODENLNCSD9.ISSN0302-9743.Garcia:1999:SI [10]D.Garcia and W.Watson.ServerNet II.Lecture Notes in Computer Science,1417:119–??,1999.CODEN LNCSD9.ISSN0302-9743.Lund:1999:ESS [11]C.Lund.Embedded systems standards.Lecture Notes in Computer Science,1417:137–??,1999.CODEN LNCSD9.ISSN0302-9743.Stunkel:1999:CDC [12]C.B.Stunkel.Challenges in the designof contemporary routers.Lecture Notesin Computer Science,1417:139–??,1999.CODEN LNCSD9.ISSN0302-9743.Byrd:1999:ECM [13]G.T.Byrd and M.J.Flynn.Evalu-ation of communication mechanisms ininvalidate-based shared memory multi-processors.Lecture Notes in ComputerScience,1417:159–??,1999.CODENLNCSD9.ISSN0302-9743.Dai:1999:HCW [14]D.Dai and D.K.Panda.How can wedesign better networks for DSM systems?Lecture Notes in Computer Science,1417:171–??,1999.CODEN LNCSD9.ISSN0302-9743.VonEicken:1999:IUW [15]T.Von Eicken.Integration of U-Net intoWindows/NT(invited presentation).Lec-ture Notes in Computer Science,1417:185–??,1999.CODEN LNCSD9.ISSN0302-9743.Smai:1999:DFC [16]A.-H.Smai and L.-E Thorelli.Distance-basedflow control in wormhole networks.Lecture Notes in Computer Science,1417:189–??,1999.CODEN LNCSD9.ISSN0302-9743.Silla:1999:UVC [17]F.Silla and J.Duato.On the use of vir-tual channels in networks of workstationswith irregular topology.Lecture Notesin Computer Science,1417:203–??,1999.CODEN LNCSD9.ISSN0302-9743.REFERENCES5Kesavan:1999:MSI [18]R.Kesavan and D.K.Panda.Multicas-ting on switch-based irregular networksusing multi-drop path-based multidesti-nation worms.Lecture Notes in Com-puter Science,1417:217–??,1999.CO-DEN LNCSD9.ISSN0302-9743.Patel:1999:PPT [19]C.S.Patel,S.M.Chai,S.Yalamanchili,and D.E.Schimmel.Power/performancetrade-offs for direct networks.LectureNotes in Computer Science,1417:231–??,1999.CODEN LNCSD9.ISSN0302-9743.McKenzie:1999:CDI [20]N.R.McKenzie,K.Bolding,C.Ebeling,and L.Snyder.ChaosLAN:Design andimplementation of a gigabit LAN usingchaotic routing.Lecture Notes in Com-puter Science,1417:247–??,1999.CO-DEN LNCSD9.ISSN0302-9743.Yuan:1999:DTM [21]X.Yuan,R.Gupta,and R.Melhem.Doestime-division multiplexing close the gapbetween memory and optical communi-cation speeds?Lecture Notes in Com-puter Science,1417:261–??,1999.CO-DEN LNCSD9.ISSN0302-9743.Warnakulasuriya:1999:MMB [22]S.Warnakulasuriya and T.M.Pinkston.Modeling message blocking and dead-lock in interconnection networks.LectureNotes in Computer Science,1417:275–??,1999.CODEN LNCSD9.ISSN0302-9743.Lopez:1999:RDF[23]P.Lopez,J.M.Martinez,J.Duato,andF.Petrini.On the reduction of dead-lock frequency by limiting message injec-tion in wormhole networks.Lecture Notesin Computer Science,1417:295–??,1999.CODEN LNCSD9.ISSN0302-9743.Shepherd:1999:ARI [24]D.Shepherd.ATM,reservation and IP—ATM,RIP?Lecture Notes in Com-puter Science,1483:1–??,1999.CODENLNCSD9.ISSN0302-9743.Parulkar:1999:HSP [25]G.Parulkar.High speed packet switch-ing and QoS:(A)guru’s perspective.Lec-ture Notes in Computer Science,1483:2–??,1999.CODEN LNCSD9.ISSN0302-9743.Geyer:1999:SAC [26]W.Geyer and R.Weis.A secure,account-able,and collaborative whiteboard.Lec-ture Notes in Computer Science,1483:3–??,1999.CODEN LNCSD9.ISSN0302-9743.Pfeifer:1999:MGL [27]T.Pfeifer,T.Magedanz,and S.Huebener.Mobile guide—location-aware applica-tions from the lab to the market.Lec-ture Notes in Computer Science,1483:15–??,1999.CODEN LNCSD9.ISSN0302-9743.Burger:1999:IPS [28]C.Burger,K.Rothermel,and R.Meck-lenburg.Interactive protocol simulationapplets for distance education.LectureNotes in Computer Science,1483:29–??,1999.CODEN LNCSD9.ISSN0302-9743.Ensor:1999:VTA [29]J.R.Ensor,G.U.Carraro,and J.T.Edmark.Visual techniques to accommo-REFERENCES6date varying network performance in vir-tual environments.Lecture Notes in Com-puter Science,1483:41–??,1999.CODENLNCSD9.ISSN0302-9743.Fromme:1999:ARK[30]M.Fromme,L.Grueneberg,andH.Pralle.An address resolution and keyexchange protocol for conferencing appli-cations on the Internet.Lecture Notesin Computer Science,1483:47–??,1999.CODEN LNCSD9.ISSN0302-9743.Villemur:1999:IPC [31]T.Villemur,V.Baudin,S.Owezarski,and M.Diaz.An integrated platform forcooperative teleteaching.Lecture Notesin Computer Science,1483:59–??,1999.CODEN LNCSD9.ISSN0302-9743.Trossen:1999:CCC [32]D.Trossen and K.-H S:CORBA-based conferencing service.Lec-ture Notes in Computer Science,1483:71–??,1999.CODEN LNCSD9.ISSN0302-9743.VanSinderen:1999:ATM [33]M.Van Sinderen and L.Ferreira Pires.The application of TINA in the MESHproject.Lecture Notes in ComputerScience,1483:77–??,1999.CODENLNCSD9.ISSN0302-9743.Deicke:1999:FMM [34]J.Deicke,U.Mayer, A.Knoll,andM.Glesner.Flexible multiplexing inMPEG-4systems.Lecture Notes in Com-puter Science,1483:83–??,1999.CODENLNCSD9.ISSN0302-9743.Kunkelmann:1999:VEB [35]T.Kunkelmann and U.Horn.Video en-cryption based on data partitioning andscalable coding—A comparison.Lec-ture Notes in Computer Science,1483:95–??,1999.CODEN LNCSD9.ISSN0302-9743.Lagha:1999:AIM [36]gha and H.Afifi.An architec-ture for an interactive multimedia sys-tem based on MPEG-2.Lecture Notesin Computer Science,1483:107–??,1999.CODEN LNCSD9.ISSN0302-9743.Wang:1999:CWW [37]J.Z.Wang,J.Li,G.Wiederhold,andO.Firschein.Coding for WWW,wireless,and mobile environments classifying ob-jectionable Websites based on image con-tent.Lecture Notes in Computer Science,1483:113–??,1999.CODEN LNCSD9.ISSN0302-9743.Melih:1999:IPC [38]K.Melih and R.Gonzalez.Identify-ing perceptually congruent structures foraudio retrieval.Lecture Notes in Com-puter Science,1483:125–??,1999.CO-DEN LNCSD9.ISSN0302-9743.Chang:1999:ICR [39]E.Y.Chang.An image coding and recon-struction scheme for mobile computing.Lecture Notes in Computer Science,1483:137–??,1999.CODEN LNCSD9.ISSN0302-9743.Iren:1999:NCI [40]S.Iren,P. D.Amev,and P.T.Con-work-conscious compressed im-ages over wireless networks.Lecture Notesin Computer Science,1483:149–??,1999.CODEN LNCSD9.ISSN0302-9743.Ingvaldsen:1999:SDF [41]T.Ingvaldsen, E.Klovning,andM.Wilkins.A study of delay factorsREFERENCES7in CSCW applications and their im-portance.Lecture Notes in ComputerScience,1483:159–??,1999.CODENLNCSD9.ISSN0302-9743.Boyer:1999:DQR [42]M.Boyer,P.Owezarski,and M.Diaz.Dy-namic QoS renegotiation in the PNSVSvideoconferencing application.LectureNotes in Computer Science,1483:171–??,1999.CODEN LNCSD9.ISSN0302-9743.Aagedal:1999:TOO [43]J.O.Aagedal.Towards an ODP-compliant object definition language withQoS-support.Lecture Notes in Com-puter Science,1483:183–??,1999.CO-DEN LNCSD9.ISSN0302-9743.Cho:1999:DGI [44]S.Cho and Y.Shin.DA VIC goes toInternet:Multimedia service interwork-ing over heterogeneous networking envi-ronment.Lecture Notes in ComputerScience,1483:195–??,1999.CODENLNCSD9.ISSN0302-9743.Tsai:1999:TFC[45]W.K.Tsai,L. C.Hu,and Y.Kim.A temporal-spatialflow control protocolfor ABR in integrated networks.LectureNotes in Computer Science,1483:207–??,1999.CODEN LNCSD9.ISSN0302-9743.Martinez:1999:LCC[46]J.Martinez,J.R.Vidal,and L.Guijacro.A low complexity congestion control algo-rithm for the ABR class of service.Lec-ture Notes in Computer Science,1483:219–??,1999.CODEN LNCSD9.ISSN0302-9743.Helbig:1999:PBC [47]T.Helbig and O.Schreyer.Protocol forbrowsing in continuous data for coop-erative multi-server and multi-client ap-plications.Lecture Notes in ComputerScience,1483:231–??,1999.CODENLNCSD9.ISSN0302-9743.Baier:1999:IDD [48]R.Baier.Implementation of a DSM-CC-server for a DA VIC-terminal.LectureNotes in Computer Science,1483:237–??,1999.CODEN LNCSD9.ISSN0302-9743.Tsirikos:1999:CDI [49]D.Tsirikos,T.Markousis,Y.Mouroulis,and M.Hatzopoulos.A client-serverdesign for interactive multimedia docu-ments based on Java.Lecture Notesin Computer Science,1483:248–??,1999.CODEN LNCSD9.ISSN0302-9743.Benz:1999:ARS [50]H.Benz and M. E.Lijding.Asyn-chronously replicated shared workspacesfor a multi-media annotation service overInternet.Lecture Notes in ComputerScience,1483:260–??,1999.CODENLNCSD9.ISSN0302-9743.Brelot:1999:OGP[51]M.Brelot and G.Privat.Object graphs asa pivotal representation for hypermedia.Lecture Notes in Computer Science,1483:272–??,1999.CODEN LNCSD9.ISSN0302-9743.Chang:1999:NRD [52]R.-I.Chang,W.-K.Shih,and R.-CChang.A new real-time disk schedul-ing algorithm and its application todistributed multimedia storage systems.REFERENCES8 Lecture Notes in Computer Science,1483:278–??,1999.CODEN LNCSD9.ISSN0302-9743.Boulos:1999:CDM[53]J.Boulos and K.Oho.Continuous datamanagement on tape-based tertiary stor-age systems.Lecture Notes in Com-puter Science,1483:290–??,1999.CO-DEN LNCSD9.ISSN0302-9743.El-Saddik:1999:EUB[54]A.El-Saddik,C.Griwodz,and R.Stein-metz.Exploiting user behaviour inprefetching WWW documents.LectureNotes in Computer Science,1483:302–??,1999.CODEN LNCSD9.ISSN0302-9743.Garcia-Martinez:1999:SPB[55]A.Garcia-Martinez,J.Fernandez-Conde,and A.Vi˜n a.Single pair of buffers:Reducing memory requirements in VBRmedia servers.Lecture Notes in Com-puter Science,1483:312–??,1999.CO-DEN LNCSD9.ISSN0302-9743.。
第1章粒计算的艺术-theDepartmentofComputerScience-University
第1章粒计算的艺术姚一豫 (Yiyu Yao)Department of Computer Science, University of ReginaRegina, Saskatchewan, Canada, S4S 0A2E-mail: yyao@cs.uregina.cahttp://www2.cs.uregina.ca/~yyao/1.1引言粒计算（Granular Computing）是一门飞速发展的新学科。
它融合了粗糙集、模糊集以及人工智能等多种理论的研究成果。
在短短十年的发展中，我们已经见证了它对科学及计算机科学的作用和影响。
诸多学者就粒计算的基本理论和方法做了大量工作（见本章参考文献），但为粒计算下一个正式的、精确的、并且能够广为接受的定义仍然是一件困难的事情。
虽然如此，我们仍然可以从问题求解及实践中提取出一些通用的理论和基本要素[1]。
我们对粒计算的描述是建立在对它的直觉认识上的：粒计算是研究基于多层次粒结构的思维方式、问题求解方法、信息处理模式，及其相关理论、技术和工具的学科。
在中国，粒计算的研究已引起众多学者的关注与兴趣。
本书的附录比较全面地收录了近年在国内期刊发表的粒计算方面的文章。
包括，基于商空间理论的粒计算模型[2]，模糊商空间及粒计算的商闭包空间模型(张钹和张铃等) [3,4,5,6]；粒计算的覆盖模型，粗糙集与粒计算的交叉问题的研究(张文修等)[7,8]；粒、规则与例外的关系(王珏等) [9,10,11,12]；粒计算的理论、模型与方法的探讨(苗夺谦等) [13,14,15,16,17,18]；基于Dempster-Shafer理论和粗糙集的近似和知识约简(吴伟志等) [19, 20,21,22]；几种基于覆盖粗糙集的粒计算模型(祝峰和王飞跃)[23,24,25]；粒逻辑及其归结原理(刘清等) [26,27,28,29,30]；基于关系的粒计算模型，粒化思想在图像的纹理识别上的应用(史忠植等) [31,32,33,34]；基于相容关系的粒计算模型，粒计算在进化计算、机器学习中的应用(王国胤等) [35,36,37,38,39]；使用粒计算进行知识获取的方法(梁吉业和李德玉) [40]；基于泛系理论的粒计算模型(李永礼和林和等) [41,42,43]；使用粒分析来描述、刻画粒计算的思考(李凡长)；等等。
- 1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。
CACR Technical ReportCACR-185 March 2000Grid Computing on the Web Using the Globus Toolkit Giovanni Aloisio, Massimo Cafaro, Paolo Falabella, Carl Kesselman,Roy WilliamsMailing Address: CACR Technical Publications, California Institute of Technology, Mail Code 158-79, Pasadena, CA 91125. Phone: (626) 395-6953 Fax: (626) 584-5917 2000 California Institute of Technology, Center for Advanced Computing Research.All rights reserved.Lecture Notes in Computer Science 1 Grid Computing on the Web Using the Globus Toolkit Giovanni Aloisio1, Massimo Cafaro1, Paolo Falabella1,Carl Kesselman2, and Roy Williams31 Dept. of Innovation Engineering,University of Lecce, Italy{giovanni.aloisio,massimo.cafaro,paolo.falabella}@unile.it2 Institute of Science Information,University of Southern Californiacarl@3 Center for Advanced Computing Research,California Institute of Technologyroy@Abstract. In this paper we present and discuss an architecture that allows trans-parent access to remote supercomputing facilities from a web gateway. The im-plementation exploits the Globus toolkit and provide users with fast, secure andreliable access to parallel applications. We show the usefulness of our approachin the context of Digital Puglia, an active digital library of remote sensing digi-tal data.1 IntroductionThis paper describes a design pattern that enables a user to run applications on a Computational Grid [1] using a standard Java-enabled browser. We define a Grid application as consisting of one or more programs, located on geographically distrib-uted machines; these programs access data stored on distributed databases. Among the most important requirements is the capability to find the right programs and to run them with the right data by means of high level requests. The user should be also al-lowed to check the status of the remote execution, get the output of the application and, in case of an interactive application, steer it from the browser.In order to integrate Grid applications and the Web, we advise the use of a three tier architecture: (1) the client web browser, (2) middleware running on a secure web-server, (3) the supercomputers of the Computational Grid.The issues related to Computational Grids and metacomputing have been abun-dantly described in literature and, in our context, we address essentially the same problems: how do we locate resources, how do we secure these resources and the transactions involving them, how do we start executing an application on a remote machine and so forth. Instruments to help solving these problems have been developed in recent years [2]; among them we chose to use the Globus toolkit which by this time has been released in its version 1.1. The rest of the paper is organized as follows. Section 2 presents the issues related to Grid applications; we discuss them in sectionsThe work presented in this paper was supported in part by ISUFI (Istituto Superiore Universi-tario di Formazione Interdisciplinare) and by CACR (Center for Advanced Computing Re-searchLecture Notes in Computer Science 23 – 7 describing the solutions provided by our framework. Section 8 shows how we exploit the framework in the context of Digital Puglia, an active digital library of remote sensing data and section 9 concludes this work.2 Web – Grid IntegrationIn this section we discuss the issues arising from the integration of Computational Grid applications with the Web; in particular, here we address the following:1. Security: The Globus tools use X509v3 certificates to mutually authenticate a userand the Grid machines. We propose a two-step mechanism to enable user – Grid authentication from a web browser: a login and password from the browser to the web-server and a certificate from there to the Grid.2. Resource management: An application running on the Grid could exploit distrib-uted, heterogeneous resources. If we want our application to find the resources dy-namically we must store somewhere the information needed to locate the machines, files, applications and so on. This is exactly the aim of the Grid Information Ser-vice, an LDAP [3] Directory Service provided by the Globus organization. Appli-cations may need a broker to translate high level requests made by the users in low level system requests.3. Remote start and status: Commands are provided in the Globus toolkit to run inter-active applications or to submit batch jobs on a remote machine taking care of the authentication issues and to check the status of the execution. These commands can be included in CGI scripts or java servlets to be run on a web server hosted on the second tier of our proposed architecture.4. Steering: A Java applet can be used to provide the user with a user friendly inter-face to the Grid application.5. Result retrieval: The final output produced by the application is held on a remotemachine or on a collection of remote machines. The choice of an appropriate strat-egy is needed to let the user have the results on her web browser in the minimum time.In order to implement the approach proposed in this paper, the Grid has to be set up in the following way:•=T he first tier needs just a Java-enabled browser: this is the basic functionality that all workstations, laptops, and operating systems can provide, •=G lobus has to be installed on the machines on the second tier and on the third tier•=A web server must be installed on the machine on the second tier: we strongly suggest it to be a secure web serverLecture Notes in Computer Science 3Fig. 1. Accessing the power of a Computational Grid via Web3 SecurityA Computational Grid consists of a set of valuable resources like parallel supercom-puters, workstations, databases and smart instruments, so in order to prevent unauthor-ized accesses to the Grid we need a strong authentication mechanism.We propose a two step mechanism: the user must authenticate herself from the cli-ent browser to the second tier and also from the second tier to the Grid. A list of trusted users is held on the second tier, and authentication via web is done through a login and password scheme. To guarantee the safety of the password we strongly sug-gest the use of a secure web server to avoid transmitting the password in clear over an insecure channel.The second step of the authentication may depend on different security policies. We can decide to map all the users on the same Globus certificate (the “guest” ac-count) or have each trusted user store her individual certificate and private key on the second tier. If we use just one certificate and private key for all the users, it is simpler because we shall not need to create an account for all the users on every machine of the Grid and to store every user’s certificate and private key on the second tier. This approach is much easier to manage and it scales better with an increasing number of machines on the third tier; however it is risky to have many people using a common “guest” password. On the other hand, we might need to use the more complex one to one mapping if the application needs to know which user is making a certain request: for instance in case of personalized options, of different privileges, or to capture accu-rate accounting information from the supercomputing resource.Lecture Notes in Computer Science 4 Before the machine on the second tier can run any Globus command on the Grid, a Globus proxy must be created with the “grid-proxy-init” command. If the mapping is one to one, the password the user has to enter to access the system must be the PEM pass phrase that protects her private key.4 Resource ManagementApplications running in a Grid environment should be resource aware and adaptive, i.e., capable of handling heterogeneity, and able to adjust their behavior dynamically in response to changes in the Grid. When the user formulates a high-level query the application should be capable of runtime discovery of computing resources (that can be geographically distributed), resource reservation through allocation and/or co-allocation and remote execution on the available pool of machines; one or more ex-ecutables may be involved in the computation, and a dataset collected from a distrib-uted database. To achieve this goal there should be:1. A way to store and retrieve information about machines, executables and data; theGlobus organization maintains the Grid Information Service, formerly known as Meta Directory Service, which is an LDAP based directory service used to provide uniform access to structure and state information about entities composing the Grid.2. One or more Resource Brokers in charge of processing high-level requests formu-lated by the users and translating them in low level requests understandable to the Grid, if necessary exploiting the information available in the Grid Information Ser-vice.3. Grid software infrastructure tools like the Globus toolkit middleware that providesa bag of core services, including a low level scheduler API, the Grid InformationService, multimethod communication and QoS management, single sign-on and key management, remote file access and Grid status monitoring.5 Remote start and status of the executionOnce the machine on the second tier has located the executables, we can use the Globus commands or API to run it remotely. A java servlet or CGI script can be easily written to incorporate the globusrun (or globus-job-submit) command.The essential parts of this CGI script are shown below (PERL language is used in this example):#environment variables needed by Globus$ENV{HOME}="/users/myapplication";$ENV{X509_CERT_DIR}=\"/usr/local/globus/share/certificates/";#creation of a grid proxy:if we are using a one to#one mapping,certname and keyname will point to the #certificate and private key we are using for#everybody.The pwstdin option allows us to read the #PEM pass phrase from the stdin and thus to use#redirectionLecture Notes in Computer Science 5 system(grid-proxy-init-pwstdin-cert$certname-key\ $keyname<$PEM_pass_phrase);#execution of the remote programsystem(globus-job-submit$remote_machine$executable);The status of execution can be checked with the utility globus-job-status, using the job-id returned by globus-job-submit and which uniquely identifies the submitted job.The application may require command-line arguments and/or input files. An applet or web form can assist the users during the phase of input generation providing a friendly GUI that can be used to derive visually input parameters; a CGI or a servlet on the second tier will then process the request generating command line arguments, files or both.The globus-job-submit and globusrun commands both allow passing command-line arguments to the application, so it is quite straightforward to add them to the sample code snippet presented above.The globus-rcp command mimics on the Grid the functionalities provided by the standard cp and rcp Unix commands and can thus be used to transfer input files on the machines belonging to the third tier before starting the remote execution.6 SteeringNot all of the Grid applications will be batch jobs that can be submitted to a pool of Grid machines, so that steering of an interactive application is an important concern here.Let us assume that we have to develop an interactive application. We advise writing a multithreaded application in which one of the threads will be responsible for han-dling the flow of control instructions via sockets, and the use of java applets to pro-vide the users with a friendly GUI that can be used to steer the application at runtime.The interaction involves bi-directional network communication over sockets, one channel is used to transmit steering instructions, the other one to report diagnostic output about commands executed.Lecture Notes in Computer Science 6Fig. 2. The use of a redirector allows the steering of interactive application without having to change the code. The user interface is in an applet. We might have to use a steering broker to connect the applet to the application if there is no possibility to install a web server on the back-end machine.The next steps in the design depend on the possibility to install a web server on the machines belonging to the third tier. This in turn depends on the local policies in use at the different sites composing the Grid. As an example, it may not be feasible to install a web server on a supercomputer because it can compromise the security of the system. Anyway, if system managers allow it, the easiest thing to do is to have the applet come directly form the third tier to the browser.Otherwise, the applet must be installed on the middle tier web server. Since un-signed applets can’t communicate over a network connection with arbitrary machines, we can address the problem either exploiting signed applets or by means of a steering broker whose aim is to dispatch to Grid machines the commands forwarded to it from the applet and to the applet the diagnostic stream output generated by the remote ma-chines. The process of applet signing is not straightforward and differs considerably if the applet is to be signed for use with different client browsers, so we feel confident that the steering broker approach is to be preferred.Communication between the steering broker and the machines on the Grid may use XTI, TLI or the Nexus communication library instead of sockets while communication with the applet is restricted to sockets due to java limits.Now, let us suppose that we have at our disposal an interactive legacy application we want to make available on the Web using the steering mechanism described above. In addition to the steering broker we need to add a redirector, that is, a small piece of code that runs on the same machine as our legacy application with the purpose of redirecting the standard input, output and error of the application to sockets connected to the steering broker connected to the steering broker (fig 2).Lecture Notes in Computer Science 7 7 Result retrievalThe final output produced by the application is held on a remote machine or on a collection of remote machines. Two strategies are possible here. We can choose either to retrieve the result directly from the machines of the third tier or to gather the results on the middle tier web server. It is difficult to predict which option will give the best results in terms of faster file transfers lacking dynamic information about the speed of network connections and the available bandwidth that may be reduced due to heavy traffic. We advise the use of a dynamic reconfiguration procedure to choose the route that minimizes the time needed by file transfers.If accelerating data output movement is not a critical issue, the designers may de-cide to trade the benefits arising from a dynamic reconfiguration with the ease of a static design, in which one of the available options is simply chosen randomly.If the output has to be retrieved directly from the third tier, Globus provides a use-ful tool, i.e., the globus-gass-server, that can be exploited to transfer files using the HTTP protocol directly from the client browser. Otherwise, if the output has to be migrated on the second tier, we suggest the use of a manager – worker scheme, desig-nating the machine on the middle tier as the manager and the others belonging to the Grid as the workers.The manager machine is in charge of gathering the output generated by the appli-cation, so that the user will retrieve all the files from the manager machine; the task can be carried out using the globus-rcp command to put the output files on a directory accessible from the web.8 Web access to the Grid in Digital PugliaDigital Puglia [4-7] is an active digital library of remote sensing data which allows interactive browsing and parallel post processing on the Grid using the Web as a gateway. A set of trusted user was defined and we provided them with a couple of parallel applications, a supervised bayesan classifier and a Principal Component Analysis.Since our users are remote sensing experts, not computer scientists, we decided to map all of them to a single Globus certificate. The users access Grid applications from the University of Lecce Digital Puglia web site, filling in an authentication form. Then, an applet allows them to interactively derive the input parameters, as needed by the applications. A couple of CGI scripts start and monitor execution progress on the remote machine, an HP Exemplar machine at Caltech.Lecture Notes in Computer Science 8Fig. 3. In the Digital Puglia active library, presented at SuperComputing’99 the possibility to access a worldwide Computational Grid via a web browser was demonstrated The final output, which is an image, is then retrieved directly from the browser where it can be visualized and saved.An example session with the supervised classifier was shown in the NPACI exhibit at SuperComputing 1999 to demonstrate the feasibility of interactive web access to Grid applications.9 ConclusionsWe have presented and discussed a framework to build Grid applications that can be transparently started and steered from a Java-enabled web browser. The architecture is based on the Globus toolkit to provide users with fast, secure and reliable access to applications on the Computational Grid. We have showed an example of the use of this framework in the context of Digital Puglia, an active digital library of remote sensing digital data.References1. Ian Foster, Carl Kesselman: The Grid. Blueprint for a new computing infrastructure.2.3.4.5.ing Environment”, Proc. HPCN Europe 1997, Vienna, Austria, Lecture Notes In Com-puter Science, Springer-Verlag, n.1225 (1997) 480-486.Lecture Notes in Computer Science 9 6. G. Aloisio, G. Milillo, R.D. Williams, ”An XML Architecture for High Performance Web-Based Analysis of Remote Sensing Archives”, Future Generation Comp. Sys., 16 (1999) 91-100.7. G. Aloisio, M. Cafaro, R. Williams, “ The Digital Puglia Project: an active digital libraryof remote sensing data”, Lect. Notes in Comp.Sci. (Springer) 1593 (1999) 563-572。