Associate Professor · AI Research Leader
Making agentic AI robust, reliable, and deployable, from the frontier to the edge.
I build AI systems that hold up when it matters: agentic systems that know when they might be wrong, models compressed to run at the edge, and rigorous methods for telling a real result from a lucky one.
Research thrusts
Robust & Reliable Visual Agentic Systems
Agents that perceive and act, and know when they might be wrong.
Efficient AI at the Edge
Compress capable models to run under real latency and power limits.
Uncertainty, Assurance & Trust
Rigorous measurement: telling a real result from a lucky one.
Current work
News
All news- Oct 2026
Chairing the LEAF workshop at IEEE MILCOM 2026
I'm in the Washington, DC area co-chairing LEAF (Lightweight Edge-Adaptive Foundation Models for Intelligence and Autonomous Reasoning), Workshop 4 at IEEE MILCOM 2026, with Venkat Dasari and Peng Wang of DEVCOM ARL. The program has eleven papers on compressing and deploying foundation models at the tactical edge, including two posters from our group: GOE and Measure, Then Compress.
- Sep 2026
Accepted to the NeurIPS 2026 AI for Science Workshop
Venkat Dasari and I will present “Your Research Loop Is Measuring Noise: Accept Rules for Autonomous Experimentation” as a poster at the AI for Science Workshop at NeurIPS 2026 in Sydney, Australia, on December 12. The work asks when an AI agent running its own experiments should trust an improvement.
- Sep 2026
CUSP preprint on arXiv
Johnny Yu, David Garcia, and I released CUSP, which splits a multi-agent vision-language system's uncertainty into each agent's own doubt and the disagreement between agents, so the system can catch errors and know when to abstain.
Work with me