Why look at case studies? It is the quickest, easiest way to gain intuition, by writing and read other’s code and architecture.

Why CNN? Inputs can be really big! For example, 1000 x 1000 becomes million.

Error Analysis Look at dev Examples to evaluate ideas.(ex. cat classifier) Should you try to make your cat classifer do better on dogs?

Example methods to imrove your model Collect more data Collect more diverse training set Train longer Try Adam instead of Gradient Descent Try bigger network Try smaller network Try dropout Add L2 regularization Network Architecture Orthogonalization Orthogonalization means to tune only one thing to achieve one effect!