NeuralNetwork(神經(jīng)網(wǎng)絡(luò)自學(xué)的英文材料)

1、1Introduction2Course Objectives This course gives an introduction to basic neural network architectures and learning rules. Emphasis is placed on the mathematical analysis of these networks, on methods of training them and on their application to practical engineering problems in such areas as patte

2、rn recognition, signal processing and control systems.3What Will Not Be Covered Review of all architectures and learning rules Implementation VLSI Optical Parallel Computers Biology Psychology4Historical Sketch Pre-1940: von Hemholtz, Mach, Pavlov, etc. General theories of learning, vision, conditio

3、ning No specific mathematical models of neuron operation 1940s: Hebb, McCulloch and Pitts Mechanism for learning in biological neurons Neural-like networks can compute any arithmetic function 1950s: Rosenblatt, Widrow and Hoff First practical networks and learning rules 1960s: Minsky and Papert Demo

4、nstrated limitations of existing neural networks, new learning algorithms are not forthcoming, some research suspended 1970s: Amari, Anderson, Fukushima, Grossberg, Kohonen Progress continues, although at a slower pace 1980s: Grossberg, Hopfield, Kohonen, Rumelhart, etc. Important new developments c

5、ause a resurgence in the field5Applications Aerospace High performance aircraft autopilots, flight path simulations, aircraft control systems, autopilot enhancements, aircraft component simulations, aircraft component fault detectors Automotive Automobile automatic guidance systems, warranty activit

6、y analyzers Banking Check and other document readers, credit application evaluators Defense Weapon steering, target tracking, object discrimination, facial recognition, new kinds of sensors, sonar, radar and image signal processing including data compression, feature extraction and noise suppression

7、, signal/image identification Electronics Code sequence prediction, integrated circuit chip layout, process control, chip failure analysis, machine vision, voice synthesis, nonlinear modeling6Applications Financial Real estate appraisal, loan advisor, mortgage screening, corporate bond rating, credi

8、t line use analysis, portfolio trading program, corporate financial analysis, currency price prediction Manufacturing Manufacturing process control, product design and analysis, process and machine diagnosis, real-time particle identification, visual quality inspection systems, beer testing, welding

9、 quality analysis, paper quality prediction, computer chip quality analysis, analysis of grinding operations, chemical product design analysis, machine maintenance analysis, project bidding, planning and management, dynamic modeling of chemical process systems Medical Breast cancer cell analysis, EE

10、G and ECG analysis, prosthesis design, optimization of transplant times, hospital expense reduction, hospital quality improvement, emergency room test advisement7Applications Robotics Trajectory control, forklift robot, manipulator controllers, vision systems Speech Speech recognition, speech compre

11、ssion, vowel classification, text to speech synthesis Securities Market analysis, automatic bond rating, stock trading advisory systems Telecommunications Image and data compression, automated information services, real-time translation of spoken language, customer payment processing systems Transpo

12、rtation Truck brake diagnosis systems, vehicle scheduling, routing systems8Biology Neurons respond slowly 10-3 s compared to 10-9 s for electrical circuits The brain uses massively parallel computation 1011 neurons in the brain 104 connections per neuron 9Neuron ModelandNetwork Architectures10Single

13、-Input Neuron11Transfer Functions12Transfer Functions13Multiple-Input NeuronAbreviated Notation14Layer of Neurons15Abbreviated NotationWw1 1,w1 2, w1 R,w2 1,w2 2, w2 R,wS 1,wS 2, wS R,=b12S=bbbpp1p2pR=aa1a2aS=16Multilayer Network17Abreviated NotationHidden LayersOutput Layer18Delays and Integrators1

14、9Recurrent Networka2 satlins Wa1 b+=a1 satlins Wa0 b+satlins Wpb+=20AnIllustrativeExample21Apple/Banana Sorter22Prototype Vectorspshapetextureweight=p2111=Prototype BananaPrototype AppleShape: 1 : round ; -1 : elipticalTexture: 1 : smooth ; -1 : roughWeight: 1 : 1 lb. ; -1 : 1 lb.MeasurementVectorp1

15、111=23Perceptron24Two-Input Caseahardlims n hardlims1 2p2+=w1 1,1=w1 2,2=Wpb+0=1 2p2+0=Decision Boundary25Apple/Banana Exampleahardlimsw1 1,w1 2,w1 3,p1p2p3b+=The decision boundary shouldseparate the prototype vectors.p10=10 0p1p2p30+0=The weight vector should be orthogonal to the decision boundary,

16、 and should point in the direction of the vector which should produce an output of 1. The bias determines the position of the boundary26Testing the Networkahardlims10 01110+1 banana=Banana:Apple:ahardlims10 01110+1 apple=“Rough” Banana:ahardlims10 01110+1 banana=27Hamming Network28Feedforward LayerF

17、or Banana/Apple RecognitionW1p1Tp2T111111=b1RR33=a1W1pb1+p1Tp2Tp33+p1Tp3+p2Tp3+=S2=29Recurrent LayerW211=1S1- - - - - - -a2t1+poslin11a2t poslina12t a22t a22t a12t =30Hamming Operationp111=Input (Rough Banana)a11111 1111133+13+13+42=First Layer31Hamming Operationa21 poslin W2a20 poslin10.50.5142posl

18、in3030=a22 poslin W2a21 poslin10.50.5130poslin31.530=Second Layer32Hopfield Network33Apple/Banana ProblemW1.20000.2000 0.2b,00.90.9=a1t1+satlins 1.2a1t =a2t1+satlins 0.2a2t 0.9+=a3t1+satlins 0.2a3t 0.9=a0 111=a1 10.71=a2 111=a3 111=Test: “Rough” Banana(Banana)34Summary Perceptron Feedforward Network

19、 Linear Decision Boundary One Neuron for Each Decision Hamming Network Competitive Network First Layer Pattern Matching (Inner Product) Second Layer Competition (Winner-Take-All) # Neurons = # Prototype Patterns Hopfield Network Dynamic Associative Memory Network Network Output Converges to a Protot

20、ype Pattern # Neurons = # Elements in each Prototype Pattern35Perceptron Learning Rule36Learning Rulesp1t1, p2t2, pQtQ, Supervised LearningNetwork is provided with a set of examplesof proper network behavior (inputs/targets) Reinforcement LearningNetwork is only provided with a grade, or score,which

21、 indicates network performance Unsupervised LearningOnly network inputs are available to the learningalgorithm. Network learns to categorize (cluster)the inputs.37Perceptron ArchitectureWw1 1,w1 2, w1 R,w2 1,w2 2, w2 R,wS 1,wS 2, wS R,=wiwi 1,wi 2,wi R,=WwT1wT2wTS=aihardlim nihardlimwTipbi+=38Single

22、-Neuron PerceptronahardlimwT1pb+hardlim w1 1,p1w1 2,p2b+=w1 1,1=w1 2,1=b1=39Decision BoundarywT1pb+0=wT1pb= All points on the decision boundary have the same inner product with the weight vector. Therefore they have the same projection onto the weight vector, and they must lie on a line orthogonal t

23、o the weight vector40Example - ORp100=t10=,p201=t21=,p310=t31=,p411=t41=,41OR Solutionw10.50.5=wT1pb+0.5 0.500.5b+0.25b+0=b0.25=Weight vector should be orthogonal to the decision boundary.Pick a point on the decision boundary to find the bias.42Multiple-Neuron PerceptronEach neuron will have its own

24、 decision boundary.wTipbi+0=A single neuron can classify input vectors into two categories.A multi-neuron perceptron can classify input vectors into 2S categories.43Learning Rule Test Problemp1t1,p2t2, pQtQ,p112=t11=,p212=t20=,p301=t30=,44Starting Pointw11.00.8=Present p1 to the network:ahardlimwT1p

25、1hardlim1.00.812=ahardlim0.60=Random initial weight:Incorrect Classification.45Tentative Learning Rule Set 1w to p1 Not stable Add p1 to 1wIf t1 and a0, then w1neww1oldp+=w1neww1oldp1+1.00.812+2.01.2=Tentative Rule:46Second Input VectorIf t0 and a1, then w1neww1oldp=ahardlimwT1p2hardlim2.0 1.212=aha

26、rdlim 0.41=(Incorrect Classification)Modification to Rule:w1neww1oldp22.01.2123.00.8=47Third Input VectorPatterns are now correctly classified.ahardlimwT1p3hardlim3.00.801=ahardlim 0.81=(Incorrect Classification)w1neww1oldp33.00.8013.00.2=If ta, then w1neww1o ld.=48Unified Learning RuleIf t1 and a0, then w1neww1oldp+=If t0 and a1, then w1n eww1oldp=If ta, then w1neww1old=eta=If e1, then w1neww1oldp+= =If e1, then w1neww1oldp=If e0, then w1neww1old=w1neww1oldep+w1oldtap+=bnewbold