HTTP/2 200 server: GitHub.com content-type: application/xml last-modified: Mon, 25 Aug 2025 18:59:37 GMT access-control-allow-origin: * strict-transport-security: max-age=31556952 etag: W/"68acb299-a1e" expires: Mon, 29 Dec 2025 15:42:03 GMT cache-control: max-age=600 content-encoding: gzip x-proxy-cache: MISS x-github-request-id: D7CF:3A7A40:8FCD1A:A161E0:69529EF3 accept-ranges: bytes age: 0 date: Mon, 29 Dec 2025 15:32:03 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210093-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1767022324.741771,VS0,VE215 vary: Accept-Encoding x-fastly-request-id: b445abbd6cd22f199cf344d10775405b5e27e7a0 content-length: 1051 Jekyll2025-08-25T11:59:18-07:00https://raonikitha.github.io/feed.xmlResearch Scientist at Google DeepMindNikitha Raoraonikitha@google.com2020-06-29T00:00:00-07:002020-06-29T00:00:00-07:00https://raonikitha.github.io/academic-posts/2020/06/blog-post-0It is easier for someone to identify Picasso’s painting than drawing one. Generative models, which are known for generating or creating data, are considered much more difficult to build and train when compared to discriminative models, which are known to process the data. We will understand the working of GANs and discuss the process involved in training. We will also try to understand why training a GAN is hard and how many of the GAN models suffer from major problems such as non convergence, mode collapse, diminished gradient to name a few. We then address the problem of mode collapse and look at some of the ways in which we can overcome this problem. This is followed by an in-depth discussion on the working of PacGAN and Unrolled GANS as solutions to the mode collapse problem.

A detailed report can be found here

]]>Nikitha Raoraonikitha@google.com