Reading tips: lecture notes
Can the network generelize to examples that have never been seen?
What is meant by overfitting?
How can overfitting be avoided?
What is meant by cross validation?