On-Policy Distillation is Reinforcement Learning in Disguise
A mathematical deep dive into why knowledge distillation, done right, is secretly policy gradient training
From REINFORCE to GRPO
[WIP] This blog is an aggregate of various papers, blogs, and videos I have been going ...
Go Proverbs
A while ago, I was watching Dwyrin's proverb series on YouTube, and to be honest, it was quite fun to watch. Since then, I have always wanted to d...
My Gate Journey
After completing my Bachelor's, I suddenly decided, without any clear reason, to prepare for GATE instead of going abroad. I had always dreamed of ...
A New Recipe for Jekyll Comments
I've always wanted to implement a custom comment box in my Jekyll blog without relying on third-party services like Disqus or the GitHub API. Since...
Replicating AlphaGo
Go, also known as Baduk in Korea, has a long history spanning over 2500 years and is loved by many in East Asia. In this game, players take turns p...
Google Summer of Code'23 - ML4SCI
I'm thrilled to share that I've been selected for Google Summer of Code (GSoC) at Ml4SCI. I'll be working on developing equivariant neural networks...