Description
Note: All coding problems to be submited with Github Link. Do not Upload the files/folder. Use git commands only.
Note: this is the distribution of questions:
-
Question 1 to Question 3: Required for everyone.
-
Question 4 to Question 5: Bonus question for both Graduate Students and Undergraduate Students
Problem 1 (10 points)
For each of the following norms, explain what properties will they favor when used in reconstruction error: L0, L1, and L2
Problem 2 (30 points)
Given a set of contrast images with sharp geometric edges (e.g. photo-lithography masks for microprocessor manufacturing) write down a formulation for recon-struction error that would work best. Justify your choice.
Problem 3 (20 points)
Given a set of images of wild life taken in their natural habitat write down a formulation for reconstruction error that would work best. Justify your choice.
Bonus for both undergraduates and gradu-
ates beyond this line.
Problem 4 (40 points)
Given distributions p and q. If q is parameterized by θ, how would you choose the value for θ to make q closest to p among all possible q’s.
-
Write down formulation of how would you measure the closeness of q to p.
-
Explain what you would do to maximize this closeness (i.e. make q and p maximally close, or minimally different or divergent)
Write a report on one of the following topics related to GANS:
-
InfoGAN https://arxiv.org/abs/1606.03657
-
CycleGAN https://arxiv.org/pdf/1703.10593.pdf
2