Currently doing a postdoc at Carnegie Mellon with Prof. Ramayya Krishnan and the AIMSEC lab, where I am pursuing the intersection of explainability, robustness, and governance of AI agents and agent systems. Recently finished my PhD at Singapore Management University, under Pradeep Varakantham and affiliated with the CARE.ai lab. My research concerns adversarial robustness in reinforcement learning and evaluating agent behavior, and I am always thinking about RL fundamentals, generalization, and AI safety. Lately I've been working on adversarial agents and benchmark evolution with the lovely folks at CyLab. I also enjoy exercising, playing games, eating, and being in the ocean.
We train LLMs to autonomously discover toxicity vulnerabilities in target LLMs through natural dialogue. We employ a hierarchical setup: one policy suggests a strategy and a second policy generates adversarial text according to the strategy. We achieve SOTA attacks on standard datasets, in addition to providing the first principled RL framework in this domain.
We progress the formulation of observation-adversarial RL by recognizing its true structure—a POMDP. Leveraging this fact, our proposed methods achieve SOTA performance across all adversarial RL benchmarks.
By optimizing a novel form of regret, we train RL agents that are more robust than previous robustly trained value-optimizing agents. Our regret notion, CCER, provides a scalable, transferrable way to compute adversarial cumulative regret for actions across time steps.
I am also proud to have participated as reviewer at the following venues: ICLR 2025, AAAI 2026, AAAI 2026: AI for Social Impact Track, ICLR 2026, ACL 2026, ICAPS 2026, NeurIPS 2026, ICLR 2027, AAAI 2027, AAAI 2027: AI for Social Impact Track, IEEE Transactions on Dependable and Secure Computing (Journal).
Presidential Doctoral Fellowship recipient, AY2024-25
Ph.D. Data Scientist Intern (Risk Research), Singapore
Associate Software Engineer, Lake Forest, California
Machine Learning Engineer, Fullerton, California
I love my work, but few are surprised to hear my true passion is the ocean. I try to schedule 1-2 dive trips per year, and I have aspirations to take and share some high-quality underwater photographs. Until then, here are a few favorites that I've taken:
Essays from my two Substacks, Roman Belaire and Take Back Your Data, mostly about the economics, social effects, and socioeconomics of AI.
Loading posts…
If you want to collaborate, have a question, or just want to say hi, you can contact me through the links below.
Pittsburgh, Pennsylvania
romanbelaire at gmail dot com
© Roman Belaire. All rights reserved. · Design derived from: HTML5 UP