Construct an ANN with 2 input nodes, 2 hidden nodes and one
output node.
Use binary inputs - so there are 4 possible input vectors.
Train it on the XOR rule so that for these
Inputs there are these Target outputs
00 0
01 1
10 1
11 0
You will be training on all
the 4 possible input sets here, do not keep any back for a test set.
Backpropagation is one obvious way to train the weights -- and you will find discussion of using backprop for this simple example in several Neural Networks textbooks. Actually there are other possiblities also, including using a Genetic Algorithm.
Bring to the seminar printouts of your code (or: legibly written pseudo-code).
Bring a pen - ideally a red one as well as a blue one! At some stage
in the seminar your efforts will be passed around, and each person will
be asked to assess, comment on and give feedback on each others work.