Learning Neural Networks: Perceptron and Backpropagation
Toets
Two main forms of learning
The Perceptron by Frank Rosenblatt (1958, 1962)
Very simple example
Learning problem to be solved
Answer
Perceptron algorithm in words
Perceptron algorithm in rules
Perceptron algorithm as equation
Perceptron algorithm in pseudo-code
Perceptron convergence theorem
The Perceptron was a big hit
Limitations of the Perceptron
Only binary input-output values
Only two layers
Exclusive OR (XOR)
An extra layer is necessary to represent the XOR
Minsky and Papert book caused the `first wave' to die out
Error-backpropagation
Error-backpropagation by Rumelhart, Hinton, and Williams
The problem to be solved
The backprop trick
Characteristics of backpropagation
The gradient descent makes sense mathematically
Logistic function
Backpropagation algorithm in rules
Weight change and momentum
Backpropagation in equations I
Backpropagation in equations II
Backpropagation in equations III
NetTalk: Backpropagation's `killer-app'
Despite its popularity backpropagation has some disadvantages
Good points
Conclusion
Opdrachten
Email: jaap@murre.com
Home Page: http://www.neuromod.org/courses/public.html
Other information: neuroMod: Home of the Neural and Cognitive Modeling Group at the University of Amsterdam.
Download presentation source
University of Amsterdam
