You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Slides http://tensorlab.cms.caltech.edu/users/anima/slides/IAS2019.pdf … New optimization: competitive gradient descent (CGD) for training GAN/multi-agent systems. Implicit competitive regularization from CGD means that we get SOTA with no explicit gradient penalty, better stability and no mode collapse #AI #DeepLearninghttps://twitter.com/RahelJhirad/status/1184134215463526401 …
.@PyTorch code for adaptive competitive gradient descent. Run it yourself to see benefits of stabilizing GANs and alleviating mode collapse. Obtained SOTA taking WGAN code with no hyperparameter tuning and no gradient penaltyhttps://github.com/devzhk/Implicit-Competitive-Regularization …
Credit goes to @Kay12400259 for working hard to build this repo
Going to read this one
Thank you!
Thank you so much for sharing :)
Massive thanks!
Wow, game changer!
Stunning work!
Great results and cool idea. The paper left me with some questions: 1) How much slower is this approach for different levels of batch size and image size? 2) What do you mean on p6 by "The additional term is a gradient step for ||nabla_y f||"? What is the norm used here?
Video available of talk at @the_IAS on using competitive gradient descent for fixing #GANs #AI on work with Florian Schaefer and @Kay12400259 at @Caltechhttps://www.youtube.com/watch?v=y4XxN3hKPDE …
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
