Reading questions

1. How does the proposed method (COPR) extend the Module network?

2. Describe the model: what are objective function, hidden/observed variables, parameters.

3. Any room for improvement, in terms of modeling or learning?