Dilip Arumugam

Welcome

I am a postdoctoral researcher in the Princeton University Computer Science Department and Psychology Department, where I work with Tom Griffiths. My research is broadly focused on reinforcement learning and, in recent years, draws heavily upon tools from information theory.

I obtained my Ph.D. from the Stanford University Computer Science Department, along with a M.S. from the Statistics Department, where I was advised by Benjamin Van Roy. I was affiliated with both the Stanford Artificial Intelligence Lab and the Stanford Computation & Cognition Lab. Prior to that, I completed B.S. and M.S. degrees in the Brown University Computer Science Department, advised by Michael Littman while also working closely with Stefanie Tellex.

Research Interests

I'm primarily interested in the area of reinforcement learning with the goal of building sequential decision-making agents that learn as efficiently and as remarkably as people do. My work employs a variety of techniques for developing principled approaches to address the core challenges of sample-efficient reinforcement learning: generalization, exploration, and credit assignment. Lately, I've been studying information theory as a collection of tools that facilitate rigorous analysis while also remaining amenable to the design of practical, scalable agents.

Selected Papers & Publications

For my CV, please click here and, for a complete list of papers, please check Google Scholar, DBLP, or Semantic Scholar (depending on what you're after, one of these may be more reliable than the others).

Bayesian Reinforcement Learning with Limited Cognitive Load

Dilip Arumugam*, Mark K. Ho*, Noah D. Goodman, Benjamin Van Roy.
Open Mind: Discoveries in Cognitive Science -- to appear, 2024.

Cultural Reinforcement Learning: A Framework for Modeling Cumulative Culture on a Limited Channel

Ben Prystawski, Dilip Arumugam, Noah D. Goodman.
Proceedings of the 45th Annual Meeting of the Cognitive Science Society (CogSci), 2023.

Deciding What to Model: Value-Equivalent Sampling for Reinforcement Learning

Dilip Arumugam & Benjamin Van Roy.
Advances in Neural Information Processing Systems (NeurIPS), 2022.
ICML Workshop on Decision Awareness in Reinforcement Learning, 2022.
Early version: Multi-disciplinary Conference on Reinforcement Learning & Decision Making (RLDM), 2022.

Planning to the Information Horizon of Bayes-Adaptive Markov Decision Processes via Epistemic State Abstraction

Dilip Arumugam & Satinder Singh.
Advances in Neural Information Processing Systems (NeurIPS), 2022.
Early version: NeurIPS Workshop on Ecological Theory of Reinforcement Learning, 2021.

The Value of Information When Deciding What to Learn

Dilip Arumugam & Benjamin Van Roy.
Advances in Neural Information Processing Systems (NeurIPS), 2021.

Deciding What to Learn: A Rate-Distortion Approach

Dilip Arumugam & Benjamin Van Roy.
International Conference on Machine Learning (ICML), 2021.

An Information-Theoretic Perspective on Credit Assignment in Reinforcement Learning

Dilip Arumugam, Peter Henderson, Pierre-Luc Bacon.
NeurIPS Workshop on Biological and Artificial Reinforcement Learning, 2020.

Value Preserving State-Action Abstractions

David Abel, Nate Umbanhowar, Khimya Khetarpal, Dilip Arumugam, Doina Precup, Michael L. Littman.
International Conference on Artificial Intelligence and Statistics (AISTATS), 2020.
Multi-disciplinary Conference on Reinforcement Learning & Decision Making (RLDM), 2019.
Early version: ICLR Workshop on Structures and Priors in Reinforcement Learning, 2019.