Udaya Ghai

I am a Senior Machine Learning Scientist in the AWS Automated Reasoning Group, working at the intersection of LLMs, reinforcement learning, and automated reasoning. Previously, I worked on applications of RL to Amazon's supply chain. More broadly, I'm interested in theory and algorithms for machine learning, particularly reinforcement learning, control, optimization (convex and non-convex), and online learning, and more recently reasoning.

I completed my Ph.D. in Computer Science at Princeton under the supervision of Prof. Elad Hazan, where we worked at the intersection of online learning and control theory. I also spent time as a student researcher at Google Brain. In a previous life, I was a sofware engineer at IMC Trading after completing a B.S. in Computer Science at Caltech.

Publications

A shared-revenue Bertrand Game Preprint.
With Raj Pabari,Dominique Perrault-Joncas, Kari Torkkola, Orit Ronen, Dhruv Madeka, Dean Foster, Omer Gottesman.
Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models ICLR 2025 (Oral)
With Yuda Song, Hanlin Zhang, Carson Eisenach, Sham M. Kakade, Dean Foster.
How Does Critical Batch Size Scale in Pre-training? ICLR 2025
With Hanlin Zhang, Depen Morwani, Nikhil Vyas, Jingfeng Wu, Difan Zou, Dean Foster, Sham M. Kakade.
Neural Coordination and Capacity Control for Inventory Management Preprint.
With Carson Eisenach, Dhruv Madeka, Kari Torkkola, Dean Foster, Sham Kakade.
Sample-Efficient Agnostic Boosting NeurIPS 2024
With Karan Singh.
Online Nonstochastic Model-Free Reinforcement Learning. NeurIPS 2023
With Arushi Gupta, Wenhan Xia, Karan Singh, Elad Hazan.
Non-convex online learning via algorithmic equivalence. NeurIPS 2022
With Zhou Lu and Elad Hazan.
A Regret Minimization Approach to Multi-Agent Control. ICML 2022 and GMAS Workshop @ ICLR 2022 (Oral Presentation + Best Poster).
With Udari Madhushani, Naomi Leonard, and Elad Hazan.
Robust Online Control with Model Misspecification. L4DC 2022
With Xinyi Chen, Elad Hazan, and Alexandre Megretski.
Machine Learning for Mechanical Ventilation Control. ML4H 2021.
With Daniel Suo, Edgar Minasyan, Paula Gradu, Xinyi Chen, Naman Agarwal, Cyril Zhang, Karan Singh, Julienne LaChance, Tom Zajdel, Manuel Schottdorf, Daniel Cohen, and Elad Hazan.
Generating Adversarial Disturbances for Controller Verification. L4DC 2021 (Oral Presentation).
With David Snyder, Anirudha Majumdar, and Elad Hazan.
Deluca—A Differentiable Control Library: Environments, Methods, and Benchmarking. DiffCVGP workshop @ NeurIPS 2020.
With Paula Gradu, John Hallman, Daniel Suo, Alex Yu, Naman Agarwal, Karan Singh, Cyril Zhang, Anirudha Majumdar, and Elad Hazan.
No-Regret Prediction in Marginally Stable Systems. COLT 2020.
With Holden Lee, Karan Singh, Cyril Zhang, and Yi Zhang.
Exponentiated Gradient Meets Gradient Descent. ALT 2020.
With Elad Hazan and Yoram Singer.