search

ritheshkumar95 / energy_based_generative_models

PyTorch code accompanying our paper on Maximum Entropy Generators for Energy-Based Models
147 stars 24 forks source link

how can we explain GAN works without I(X,Z) term? #3

Open bojone opened 5 years ago

bojone commented 5 years ago

compared with wgan-gp or wgan-div, your new GAN has an additional I(X,Z) term on generator loss. This term may prevent generator from mode collapse.

As we known, wgan-gp or wgan-div can be trained successfully without I(X,Z). But as your derivation in your paper, I(X,Z) is indispensable.

Therefore, how can we understand the success of wgan-gp or wgan-div under your framework ?

bojone commented 5 years ago

In a word, I am really interested in your model but I want to make the whole derivation more naturally.

ritheshkumar95 commented 5 years ago

I would not like for you to think about GANs (wgan-gp, wgan) in the first place. The objective of this paper is to not find a new GAN. It's to start from the theory of maximum-likelihood energy-based models and find a better way of learning / training energy-based models.

The connection to GANs is just an interesting coincidence.