You've learned a bit about what neural networks are good for, and what they're not good for. Would they be good at playing board games, like Chess, Go, or Backgammon? Would they be good at some aspects of such games, and bad at other aspects?

How would you represent neuron inputs & outputs, connection weights, and biases, in a Matlab program?



Below are some answers to the second question.

Weights: matrix. Row index = neuron index in 1 layer. Column index = neuron index in the next layer.

Biases: vector. Index = neuron index. Or as the last row of the weight matrix.

Neuron's activities: vector. Index = neuron index. To get the next layer's input: matrix-vector product.

How to turn this into matrix-matrix products? Those are more efficient. Answer: activities = matrix. Row index = neuron index. Column index = training case index.