Some students are still having trouble with plotting responses. The following piece of code provides a function which will do this for you. Just insert your own model function where I have mlpforward(...), and if you aren't interested in the multi-layer perceptron, remove the code after the first mesh().

Another point is that people are confused about the plotting options. Look at the help information on the commands plot, plot3, surf, contour, mesh for ideas. Plot is essentially for 2-D, while plot3 allows you to work with 3D plots.

Question:

OK, the main equation at the bottom of the page contains yn(1-yn). Doesn't this mean that if the output from the neuron is either 0 or 1 there will be no change in the weights? Surely if the expected output is 1 and the actual output is 0 we want to change the weights in some way but this equation won't do anything.

Answer:

well spotted! The neuron only shows plasticity (the ability to change) in the area where the derivative is non-zero, so if you have a unit which has large weights and is starting to look like a threshold function, it begins to get 'stuck'....

You might want to plot the derivative of the sigmoid function with your network results to gain some insight into this. (the y(1-y) is the derivative)

Question:

Hi There, I'm one of the Neural Computing 4 students, and I'm afraid that I've been having serious difficulty in getting to grips with Matlab. I was wondering if it would be possible to meet up for an hour or so this week to go through the previous weeks exercises, as I'm afriad I'm completely clueless as to what's going on, despite the fact that I think I'm following the lectures fine. I also wanted to query the hand-in date for the assessed exercise. You have stated in the exercise that it is to be this Friday, however Michael Jamieson explicitly stated in his Thursday lecture that none of the practical work would be due in until the end of term. I know that I am not the only person in the class having difficulties with the Matlab work, and so understandalby we are distressed at having so little time to work on the exercise, especially given our other commitments in other modules

Answer:

I'm sorry to hear that you are having trouble, but you should use the time in labs to ask about that. Michael also claims that he said no such thing - he said that the end of the lecture next Tuesday would be OK, if people had trouble making this Friday.

Also if you had kept up with the first 2 labs, this one is almost nothing new, so it is not that short notice... I'll come along to the lecture today, to see if anyone has questions