Andy (Zehao) Zhao

I am a PhD student in Electrical Engineering & Computer Science at the University of Southern California, working in the Center of Autonomy & Artificial Intelligence. I am advised by Rahul Jain.

My research studies AI agents that learn and adapt after deployment, connecting LLM post-training, multi-agent game theory, test-time adaptation, and systems optimization. I currently work with Meta Superintelligence Lab (FAIR), after research roles at Microsoft Research and Samsung Research America.

Previously, I studied at the MIT Operations Research Center with Dimitris Bertsimas. At UC Berkeley, I was the first student in the university's history to complete a quadruple major in Electrical Engineering & Computer Sciences, Statistics, Pure Mathematics, and Data Science. I worked with James Demmel, Saul Perlmutter, and Jiahao Yao.

Portrait of Andy (Zehao) Zhao

Research Highlights

My research focuses on adaptive intelligence: proactive AI agents and language models that continue to learn, anticipate, reason, and coordinate after deployment. The goal is to make adaptation persistent—new capabilities should be acquired at test time, retained across long contexts, and improved through interaction without sacrificing stability or efficiency.

Across LLM post-training, multi-agent game theory, test-time adaptation, and systems optimization, I pursue one question: how can an AI system keep getting better after it enters the world?

  • Persistent Adaptation: Building test-time learning, recursive self-improvement, on-policy distillation with black-box teachers, and long-context plasticity mechanisms that acquire new behavior while measuring retention, locality, and generalization.
  • Proactive & Strategic Agents: Developing situation-aware agents while studying bounded-rational equilibria, cooperative self-evolution, and preference optimization for agents that learn from—and against—other decision-makers.
  • Efficient Systems for Evolving Models: Co-designing post-training and inference through multi-teacher distillation, efficient KV caches, and heterogeneous compute systems that make continuous adaptation practical.

News

Research

Proactive Agent with Situation Awareness

ICML

Proactive agents that maintain situation awareness and adapt their plans as multi-agent environments and other decision-makers change.

Black-Box Teachers: On-Policy Distillation with Forward KL

ICML

On-policy distillation from black-box teachers using forward-KL objectives to improve reasoning while keeping training aligned with the student's evolving behavior.

Forecast, Act, Correct: Prediction Residual-Stability with Fallible Opponent Models

ICML

A residual-stability framework for acting against imperfect opponent models, using prediction errors to correct behavior without destabilizing the learning loop.

Hierarchical Plasticity: Test-Time Adaptation in Long-Context Transformers

ICLR

A hierarchical approach to test-time adaptation that coordinates fast local updates, working memory, and slower model changes across long contexts.

Co-evolutionary Cooperative Game for Multi-Self-Evolving AI Agents: Bounded-Context Equilibrium

AAAI

A cooperative-game formulation for populations of self-evolving agents whose policies and shared knowledge adapt under bounded context.

Multi-Agent DPO in Strategic Reasoning for LLM Post-Training

ICLR

Preference optimization for strategic multi-agent reasoning, aligning language-model policies through comparisons shaped by interacting opponents and collaborators.

Satellite map of land-release assessment areas around San Carlos, Colombia

Land Release Impact Assessment in Colombia

United Nations Open Science and Open Scholarship Conference, 2022Project reportSummarySlides

Machine-learning analysis of pre- and post-clearance satellite imagery to identify buildings, roads, and land cover and support evaluation of mine-action interventions.

Experience

USC Center for Autonomy & Artificial Intelligence2023–26
Research Assistant

Developed multi-agent reinforcement learning and post-training alignment methods, applying game theory to language-model post-training, human-AI strategic play, and multi-objective preference optimization.

Advisor: Rahul Jain

Multiple Startups (OmniGPT, FlowGPT, Ditto.AI), LLM Post-Training and Agentic AI2023–25
Founding Research Engineer

Built personalized AI systems and post-trained language models using interactions from more than 100 users and over 10,000 documents across long contexts and time horizons.

Advisor: Tianyi Peng

Ameriprise Financial2023
Data Scientist

Researched data-specific training and evaluation metrics for deep learning in investment applications.

Education

Awards & Service

Selected Honors

  • MHI Scholar Award, YC Startup School Award, OpenAI Research Credits, and Top 10 Google DeepMind Hackathon — 2026
  • MATS Research Award and Early Career Innovator Award — 2025
  • USC Conference Travel Grant and Ming Hsieh Fellowship — 2024
  • Viterbi Graduate Fellowship and Annenberg School Award — 2023

Professional Service

  • Reviewer for NeurIPS, ICLR, AISTATS, ICML, and UAI
  • Invited research talks at YCML, Cornell, MIT, Caltech, USC, and UC Irvine
  • Teaching and coursework

Contact

zhaozeha [at] usc [dot] edu  ·  zehao_zhao [at] berkeley [dot] edu

USC
EEB 335
Hughes Aircraft Electrical Engineering Center
3740 McClintock Ave.
Los Angeles, CA 90089
UC Berkeley
866 Evans Hall
Department of Mathematics
University of California, Berkeley
Berkeley, CA 94720-3840