Nilaksh

I am a first year ELLIS PhD student at Mila supervised by Prof. Sarath Chandar and co-supervised by Prof. Caglar Gulcehre. I am interested in working on RL and Robotics. I graduated from Indian Institute Of Technology, Kharagpur with a dual degree (bachelors + masters) in Electronics and Electrical Communication Engineering with a specialization in Vision and Intelligent Systems. I also got a minor in Computer Science.

Previously, I was a research intern at CLAIRE, EPFL, where I worked on multi-objective RL with Prof. Caglar Gulcehre. I was also a research intern at FRPG where I worked on multi-task adaptive RL with Prof. Aamir Ahmad. I was also fortunate to work with Prof. Shishir Kolathaya, RBCCPS IISc, on barrier function inspired reward shaping.

At IIT Kharagpur I've worked on NLP and formal language with Prof. Aritra Hazra, and on defense against SSL backdoor attacks with Prof. Somesh Kumar for my bachelor's thesis.

Email  /  CV  /  Scholar  /  Github  /  LinkedIn

Research Interests

I'm interested in reinforcement learning, computer vision, efficient deep learning and robotics. I hope to work on research problems that deeper my current understanding of these subjects, while also prompting me to think in new ways.

Publications

Defense against backdoor attacks on Self-Supervised Learning in frequency space.

A novel adaptive agent that leverages transfer learning techniques to dynamically adapt policy in response to different tasks and environmental conditions

A novel safety-oriented reward-shaping framework inspired by barrier functions, offering simplicity and ease of implementation across various environments and tasks.

Projects

A solver for SAT level geometry problems that uses a domain adapted BART transformer to predict correct geometric theorem sequences.

A simple compiler written in C++ that compiles a BASIC like language to C code. Produces the abstract syntax tree (AST) of the source for optimizations.

Simulates natural selection in a small world by mutating rudimentary genes and using a simple genetic algorithm. Conforms to the Lotka–Volterra predator–prey model.

This website's template is taken from Jon Barron.