Questions for
"Deep multi-scale video prediction beyond mean square error" by Mathieu et al.

Send your answers to tatarchm@cs.uni-freiburg.de before July 13th, 2016, 10am.

1. Why do the authors propose to use the multi-scale network? (~1-2 sentences)
2. Describe the two main components of a typical GAN (Generative Adversarial Networks) system. (~3-4 sentences)
3. Do you think using 3D convolutions instead of the 2D ones could improve the performance? Can you see any potential problems with this approach? (~2-3 sentences)