1 Pokemon/Digimon Classifier

1.1 Observation

1.2 Function

1.3 Loss

1.4 Training Examples

–

OOPS

what do we want?

2 What is the probability of sampling bad $D_{train}$

2.1 Theroy

–

Larger $N$

smaller $|H|$

2.2 Example

2.3 What if the parameters are continuous?

–

2.4 Why don’t we simply use a very small $|H|$

2.5 Tradeoff of Model Complexity