Scaling Multi-Agent Intelligence with Substrate-Centric Coordination
Zhuokai Zhao
Staff AI Research Scientist, Meta AI
PhD in Computer Science, University of Chicago
PhD in Computer Science, University of Chicago
Biography
I'm a Staff AI Research Scientist at Meta AI. Most of my work these days is on LLM agents and multi-agent systems: training agents with reinforcement learning when real-world experience is scarce (synthetic environments, experience synthesis, on-policy distillation), giving them memory and self-improvement loops for long-horizon tasks, and getting multiple agents to communicate and reason together.
Before Meta, I did my PhD in Computer Science at the University of Chicago, advised by Prof. Yuxin Chen. Earlier, I received a Master's in Robotics from Johns Hopkins University, working with Prof. Nassir Navab, and a Bachelor's with Honors in Electrical Engineering from UIUC, where I did robotics research with Prof. Seth Hutchinson.
News
-
COLM 2026 · SandMLE accepted.
Jul 2026
-
ICML 2026 · Three papers accepted: FusionRoute, LILO, and DAG-MoE.
Apr 2026
-
ACL 2026 · Mixture-of-Minds (Main) and TARo (Findings) accepted.
Apr 2026
-
ICLR 2026 · DreamGym accepted.
Jan 2026
-
AISTATS 2026 · RLMesh accepted.
Jan 2026
-
NeurIPS 2025 · Four papers accepted: Thought Communication Spotlight, Dropout Decoding, S'MoRE, and MJ-Bench.
Sep 2025
-
EMNLP 2025 · DISCO accepted.
Aug 2025
-
COLM 2025 · SPOC accepted.
Jul 2025
-
ICCV 2025 · RankCLIP and DyTo accepted.
Jun 2025
-
CVPR 2025 · HumanMM accepted; ICoV accepted to the MAR Workshop Oral.
Feb 2025
-
ICLR 2025 · SCYLLA accepted.
Jan 2025
-
NeurIPS 2024 Workshops · DAO accepted to the BDU Workshop Oral; NERO accepted to the Interpretable AI Workshop.
Oct 2024
-
CoRL 2024 · EscIRL accepted.
Sep 2024
-
IROS 2024 · ACS accepted.
Jun 2024
-
Joined Meta as a Research Scientist.
Jun 2024
-
ICML 2024 · HALC accepted.
May 2024
-
ICME 2024 · Multimodal Guidance Network accepted.
Apr 2024
-
NAACL 2024 · AutoPRM accepted.
Mar 2024
-
KDD 2023 · User-Centric Ranking accepted.
May 2023
Selected Publications
* co-first author · † joint last author
2026
SplitJEPA: Learning Invariant and Variant Latent Worlds without Reconstruction
Discovered, Not Designed: Population Evolution for Collaborative and Compute-Intensive Model Discovery
Mixture of Self-Improving Branches For Agent Harness Optimization
Rethinking Transfer in Continual Learning: A Replay-Based Realisation
Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents
SWE-Together: Evaluating Coding Agents in Interactive User Sessions
SAGE-OPD: Selective Agent-Guided Intervention for Multi-Turn On-Policy Distillation
OmniOPD: Logit-Free On-Policy Distillation via Speculative Verification
Self-Policy Distillation via Capability-Selective Subspace Projection
Self-Evolving Multi-Agent Systems via Decentralized Memory
Token-Level LLM Collaboration via FusionRoute
TARo: Token-level Adaptive Routing for LLM Test-time Alignment
Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding
Synthetic Sandbox for Training Machine Learning Engineering Agents
CircuitProbe: Tracing Visual Temporal Evidence Flow in Video Language Models
Scaling Agent Learning via Experience Synthesis
2025
Exploring System 1 and 2 Communication for Latent Reasoning in LLMs
Let it Calm: Exploratory Annealed Decoding for Verifiable Reinforcement Learning
Thought Communication in Multiagent Collaboration
Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding
StreamMem: Query-Agnostic KV Cache Memory for Streaming Video Understanding
RankCLIP: Ranking-Consistent Language-Image Pretraining
Beyond Training: Dynamic Token Merging for Zero-Shot Video Understanding
CAFe: Unifying Representation and Generation with Contrastive-Autoregressive Finetuning
Autonomous Multimodal Reasoning via Implicit Chain-of-Vision
Beyond Reward Hacking: Causal Rewards for Large Language Model Alignment
2024
Direct Acquisition Optimization for Low-Budget Active Learning
Evaluating Machine Learning Models with NERO: Non-Equivariance Revealed on Orbits
Preference Optimization with Multi-Sample Comparisons
Multimodal Guidance Network for Missing-Modality Inference in Content Moderation
PANDORA: Detailed LLM Jailbreaking via Collaborated Phishing Agents with Decomposed Reasoning
Safe Reinforcement Learning via Hierarchical Adaptive Chance-Constraint Safeguards
HALC: Object Hallucination Reduction via Adaptive Focal-Contrast Decoding
AutoPRM: Automating Procedural Supervision for Multi-Step Reasoning via Controllable Question Decomposition
2023
RELAX: Reinforcement Learning Enabled 2D-LiDAR Autonomous System for Parsimonious UAVs
Breaking the Curse of Quality Saturation with User-Centric Ranking
2020
System and Method for Assisted Patient Positioning
2026
Scaling Multi-Agent Intelligence with Substrate-Centric Coordination
SplitJEPA: Learning Invariant and Variant Latent Worlds without Reconstruction
Verify Less, Evolve More: Training Idea-Level Critics for Verification-Efficient ML Evolving Agents
Discovered, Not Designed: Population Evolution for Collaborative and Compute-Intensive Model Discovery
Mixture of Self-Improving Branches For Agent Harness Optimization
Knowledge Distillation During Mid-Training Favors Reasoning over Factual Recall
Shared Actors Need Not Share Critics: Effects of Value Mismatch in Parallel Reinforcement Learning
Rethinking Transfer in Continual Learning: A Replay-Based Realisation
SVR-R1: Bootstrapping Multi-modal Reasoning with Self-verification in Reinforcement Learning
Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents
SWE-Together: Evaluating Coding Agents in Interactive User Sessions
SAGE-OPD: Selective Agent-Guided Intervention for Multi-Turn On-Policy Distillation
OmniOPD: Logit-Free On-Policy Distillation via Speculative Verification
Self-Policy Distillation via Capability-Selective Subspace Projection
Self-Evolving Multi-Agent Systems via Decentralized Memory
Agentic Recommender System with Hierarchical Belief-State Memory
LILO: Bayesian Optimization with Interactive Natural Language Feedback
Token-Level LLM Collaboration via FusionRoute
DAG-MoE: From Simple Mixture to Structural Aggregation in Mixture-of-Experts
TARo: Token-level Adaptive Routing for LLM Test-time Alignment
Mixture-of-Minds: Multi-Agent Reinforcement Learning for Table Understanding
Synthetic Sandbox for Training Machine Learning Engineering Agents
LLM-Driven Reasoning for Constraint-Aware Feature Selection in Industrial Systems
CircuitProbe: Tracing Visual Temporal Evidence Flow in Video Language Models
Accelerating PDE Surrogates via RL-Guided Mesh Optimization
Scaling Agent Learning via Experience Synthesis
2025
Exploring System 1 and 2 Communication for Latent Reasoning in LLMs
Let it Calm: Exploratory Annealed Decoding for Verifiable Reinforcement Learning
Thought Communication in Multiagent Collaboration
S'MoRE: Structural Mixture of Residual Experts for LLM Fine-tuning
MJ-BENCH: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation?
Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding
RecoWorld: Building Simulated Environments for Agentic Recommender Systems
StreamMem: Query-Agnostic KV Cache Memory for Streaming Video Understanding
DISCO Balances the Scales: Adaptive Domain- and Difficulty-Aware Reinforcement Learning on Imbalanced Data
Boosting LLM Reasoning via Spontaneous Self-Correction
RankCLIP: Ranking-Consistent Language-Image Pretraining
Beyond Training: Dynamic Token Merging for Zero-Shot Video Understanding
Transfer between Modalities with MetaQueries
CAFe: Unifying Representation and Generation with Contrastive-Autoregressive Finetuning
HumanMM: Global Human Motion Recovery from Multi-shot Videos
Autonomous Multimodal Reasoning via Implicit Chain-of-Vision
Quantifying Generalization Complexity for Large Language Models
Beyond Reward Hacking: Causal Rewards for Large Language Model Alignment
2024
Direct Acquisition Optimization for Low-Budget Active Learning
Evaluating Machine Learning Models with NERO: Non-Equivariance Revealed on Orbits
EscIRL: Evolving Self-Contrastive IRL for Trajectory Prediction in Autonomous Driving
Preference Optimization with Multi-Sample Comparisons
Multimodal Guidance Network for Missing-Modality Inference in Content Moderation
PANDORA: Detailed LLM Jailbreaking via Collaborated Phishing Agents with Decomposed Reasoning
Safe Reinforcement Learning via Hierarchical Adaptive Chance-Constraint Safeguards
HALC: Object Hallucination Reduction via Adaptive Focal-Contrast Decoding
AutoPRM: Automating Procedural Supervision for Multi-Step Reasoning via Controllable Question Decomposition
2023
RELAX: Reinforcement Learning Enabled 2D-LiDAR Autonomous System for Parsimonious UAVs
Breaking the Curse of Quality Saturation with User-Centric Ranking
2020
Dissertations
Enhanced Data Utilization for Efficient and Trustworthy Deep Learning
Utilizing Both Past and Future: Multi-Frame Memory Based Network in Solving Particle Image Velocimetry
Other Projects
OpenChemistry/Stempy: Stable Version
Trajectory Planning and Control for Nonholonomic Robot Among Onstacles
Head-Mounted Display Integration for Orthopedic Surgery
Service
Area Chair
- Conference on Language Modeling (COLM)2026
Conference Reviewer
- International Conference on Machine Learning (ICML)2025, 2026
- Neural Information Processing Systems (NeurIPS)2023–2025
- International Conference on Learning Representations (ICLR)2023–2026
- International Conference on Computer Vision (ICCV)2025
- AAAI Conference on Artificial Intelligence (AAAI)2025, 2026
- Conference on Language Modeling (COLM)2025
- International Conference on Artificial Intelligence and Statistics (AISTATS)2025, 2026
- SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)2024
- Empirical Methods in Natural Language Processing (EMNLP)2023, 2024
- International Conference on Robotics and Automation (ICRA)2024
- International Conference on Intelligent Robots and Systems (IROS)2024
- ACM Multimedia (ACM MM)2024
- SIAM International Conference on Data Mining (SDM)2024
- IEEE International Symposium on Mixed and Augmented Reality (ISMAR)2023–2025
- IEEE Visualization Conference (IEEE VIS)2023
- International Conference on Information Systems (ICIS)2023
Journal Reviewer
- IEEE Transactions on Knowledge and Data Engineering (TKDE)2025
- IEEE Transactions on Visualization and Computer Graphics (TVCG)2023
- Transactions on Machine Learning Research (TMLR)2023