Undergraduate Researcher
Information • Games • Systems • Computational Intelligence
Email: [fname].[lname]@kgpian.iitkgp.ac.inMy interest in communication/information theory and experience running a cash-strapped undergrad robotics group have played a major role in my interest in parallel, distributed and real-time/online learning with the goal of leveraging large scale, personal computing to learn deep models in the wild.
The primary motivation behind my research is developing the tools and methodology to make AI systems more usable and democratic. This involves a bidirectional push. From the bottom-up, this means large scale computation and data. From the top-down, this means figuring out how to train more efficient, generalizable and safe models.
This requires a study of a wide range of interesting topics, from networks and distributed systems to unsupervised learning, randomized algorithms, and game theory.
Here are some of my research areas, accompanied by a brief description of some direction(s) I am (or hope to) investigating.
An asynchronous mechanism for backward feedback in energy based (or other greedy/localized) learning algorithms.
Information propagation in generative networks and privacy-preserving coding using generative networks.
A peer-to-peer fault tolerance strategy for distributed learning systems.
Inverse reinforcement learning for vehicle/pedestrian trajectory prediction in real-time.
An information-theoretic framework for analyzing convergence and complexity of randomized algorithms, particularly Las Vegas algorithms.
I have not had a chance to fill out this section yet (WIP). Other than the projects on my CV (see menu bar), I've worked on some image processing and computer vision (mainly robotic perception: scene understanding and trajectory prediction). In late '23 to early '24 I plan to take up some projects in inverse RL and unsupervised vision.