CSC321:Neural Networks

Lecture 3: Learning in multi-layer networks

Preprocessing the input vectors

The connectivity of a perceptron

Is preprocessing cheating?

What can perceptrons do?

Why connectedness is hard to compute

Learning with hidden units

Learning by perturbing weights

The idea behind backpropagation

A change of notation

Non-linear neurons with smooth derivatives

Sketch of the backpropagation algorithm
on a single training case

The derivatives

Ways to use weight derivatives

Overfitting

A simple example of overfitting