CS1674: Homework 8 - Written

1. Why do we want a classifier that gives a large margin between positive and negative examples?
2. Why can we do the "kernel trick" and not explicitly compute a lifting transformation?
3. Name the two ways to implement a multi-class classifier.
4. Describe the problem of "overfitting", and why it is tricky to handle.
5. What two things do we need, in order to enable zero-shot classification of never-seen animals?