Creator of T2F, MSG-GAN and FAGAN | Independent Researcher: AI, DL, RL, GANs | Exploratory Gamer | steamid: NranikC

Hi Valentina,

Thank you for the appreciation. The Variational Discriminator Bottleneck is a highly generic (quite seminal as well) technique, which can be used with any Adversarial Machine Learning framework. In case of image translation, you can easily incorporate these modifications in the Markovian Dsicriminator…

Hi Shamane Siriwardhana. Thank you for your comment. I am glad that my PyTorch implementation was helpful.

Best regards,

Animesh

Thank you!

Hi Soumya, I believe there is a slight confusion about the terminology of intermediate layer connections. The connections (skip connections wouldn’t be a proper term here) do not constitute any soft operation in the discriminator computations. The output of the generator at various scales is fed to the discriminator layers via concatenation. So…

Sure. Thanks!

Hi Geroge, https://buzzrobot.com/mathematical-optimization-simplicity-is-all-you-need-d0cea72b8d5a

I have explained the results of this optimization technique on Neural network this blog.

Hi! By histogram pairs, I mean pairs of histograms of the same layer for different runs. For example, pair 1 has the layer_1 biases as recorded for the adam training run and for the proposed optimizer run; similarly pair 2 has the layer_1 kernels for the same, and so on.

correction: ‘even’ instead of ‘event’

Thank you for the feedback!

As you mentioned, the BFGS algorithm uses the Hessian matrix, which is a second order derivative matrix. Although it gives good approximation of the direction to move towards for reaching the optimum value, it is computationally costlier than all the optimizers that I tested. One of the important…

This post purely describes the vision of the future. The ‘Software 2.0’ synecdoche is absolutely ingenious! I am happy that many of my views resonate with this post. However, it saddens me a bit that this vision is also being criticized. Skeptics mention that the one true Neural Network Architecture that will build towards the AGI is impossible. I…