Skip to content

Probabilistic Future Video Frame Prediction using Generative Adversarial Networks by employing a regret minimization strategy for training GANs.

Notifications You must be signed in to change notification settings

himol7/GANs-Regret-Minimization

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 

Repository files navigation

GANs-Regret-Minimization

Probabilistic future video frame prediction using Generative Adversarial Networks by employing a regret minimization strategy for training GANs.

As proposed by Ian Goodfellow in 2014, Generative Adversarial Networks are essentially two networks, a Generator and a Discriminator, pressed against eachother that learn from the opposite one's losses in a minmax game.

Generative Adversarial Networks:

alt text

Source: https://www.slideshare.net/ckmarkohchang/generative-adversarial-networks/11

minmax V(D,G) represents the min-max game between the Discriminator and Generator. Loss function used here: Log loss function.

Regret Minimization:

alt text

G: Generator D: Discriminator G-loss: Generator's loss D-loss: Discriminator's loss

For every iteration, D is supplied with G's output and D-loss is calculated. The D-loss for current iteration is propogated to next iteration where it is averaged with D-loss of that iteration and the mean loss is supplied to G for training it.

Results:

alt text

G-loss converges significantly faster when averaged with previous losses.

About

Probabilistic Future Video Frame Prediction using Generative Adversarial Networks by employing a regret minimization strategy for training GANs.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published