Neurocomputing: Fundamentals of Computational Neuroscience (Fall 2006)

Assignment 3

1. Implement a perceptron and train it to recognise the digital letters given in pattern1. Please specify your implementation of the perceptron.

(Hint: The MATLAB function patternDisp.m reads the patterns from the file pattern1 (patternDisp(n), with n=1-26) and pattern2 (patternDisp(0)) and displays them. Look at the code. It uses the Matlab function reshape that can help you to convert the patterns into input vectors. If you like you can use the delta-rule discussed in Chapter 10.2. )

2. Plot a training curve, which is the error rate as a function of training steps.

3. Investigate how much noise in the pattern the perceptron can tolerate before beeing unable to recognize a letter.

(Hint: There are different ways to implement noise, but you could just choose to flip a certain number of bits (but don't flip one twice)).

4. Which letter is in pattern2?