人工智能NN-BP_355201376

The excitation of one unit would transfer to the other, if the two units act simultaneously or locate closely. Moreover, the active level of one unit is proportional to the number and density of action around it.
Other reasons at that time: undeveloped computer, non-emerging VLSI, as well as the thriving period of Artificial Intelligence and Expert System
16
History (cont.)
6
Motor Imagery System (Gao et al.)
Foot Left Hand Right Hand
7
Dialing Phone Keypad
13Hz
15Hz
Frequency Resolution: 0.25Hz
2
Content
Introduction Single layer Perceptron Multilayer Perceptron Self-organized Networks
Self-organized mapping ART
Applications
3
Introduction
4
Outline
Hebb Law (used in Self-Organization Network learning)
If the outputs of two neurons are both excitation, connection between them is enhanced, or else reduced.
Why Neural Networks? History What are neural networks Models of a neuron Types of neural networks Characteristics Applications Neural networks and AI
5
8
History
In 1890, American biologist, W. James, published the book named “Physiology (生理 学)” which firstly indicated human brain structure and its function as well as basic laws related to learning, association, and memory. As illuminated in this book
9
History (cont.)
In 1943, McCulloch (psychologist) and Pitts (symbolic logician) presented M-P Model in which a simple artificial neuron was considered to be binary (excitation and inhibition) devices with fixed thresholds. They summarized the basic physiological features of neuron, and formalized its mathematical description and network construction. 1 Output expression: xi Neural computing period was coming!
17
History (cont.)
In1982, John J. Hopfield（physical scientist）presented fullconnected network and discrete neural network model, which was a bran-new model with complete theory basis. Its basic idea was that a specific neural network was assigned with a energy function which is proportional to the active value of each neuron and the connection weight between neurons. The active value varying algorithms aimed to reach a local minimum by energy function degression. He proved the two cases (discrete and continuous) that network would reach steady state. Three years later, AT&T etc. produced the semiconductor chip. 18 The re-emerging period was coming.
Why Neural Networks?
As part of neuroscience, in order to understand real neural systems, researchers are simulating the neural systems of simple animals such as worms. It seems reasonable to try to build the functionality of the brain via the mechanism of the brain (suitably abstracted). The brain inspires new ways to think about computation Neural networks provide a different measure of simplicity as a learning bias.
wij wij S i S j
13
History (cont.)
Hebb Law’s Contribution (present the concept of variability)
Present that information in neural networks is stored in the connection weight of synapse The learning ratio of connection weight is proportional to the product of the active state values of the two connected neural cells. Assume that weight is symmetrical The structure of cell connections is due to the variability of weights.
12
History (cont.)
In 1949, Donala U.Hebb (psychologist) published the book “The Organization of Behavior (行为自组织)”.
Present the assumption on synapse relation strength variability Indicate that learning process occurs among synapses Indicate that synapse relation strength varies with the activities of its neighbor neurons
10
y 0
x
i

Network of neurons and synapses
11
History (cont.)
M-P Model’s Contribution Accomplish some logic function Adapt collective parallel computing structure to describe artificial neuron and network initially Provide evidences for further research (achieve Boolean calculation)
y j f ( w ij x i j )
i 1 n
1 f (u j ) 0
uj 0 uj 0
15
History (cont.)
In 1969, M.Minsky and S. Papert wrote a book named “Perceptrons”, which indicated the perceptron only solved first-order Predicate Logic and linear partition. It failed to solve nonlinear or other partition problems, which was proved by a simple XOR problem. The neural network research got its frustrated period.
Chapter 6
Artificial Neural Networks