Project Highlights
Our research focuses on developing AI and machine learning methods for accelerated scientific discovery, optimization, and decision-making in complex systems.
✦ Featured Projects
LabOS: The AI-XR Co-Scientist That Sees and Works With Humans
Modern science advances fastest when thought meets action. LabOS represents the first AI co-scientist that unites computational reasoning with physical experimentation through multimodal perception, self-evolving agents, and Extended-Reality(XR)-enabled human-AI collaboration. From cancer immunotherapy to stem-cell engineering, LabOS transforms laboratories into intelligent, collaborative environments.
LabOS System Introduction
Jensen Huang GTC 2025 Keynote Feature
MATH-Perturb: Benchmarking LLMs' Math Reasoning Against Hard Perturbations
Accepted in ICML 2025. MATH-Perturb investigates whether LLMs achieve mathematical reasoning through true understanding or memorization. We construct perturbed math problems that reveal significant performance drops across various models, raising concerns about a novel form of memorization where models blindly apply learned skills without assessing their applicability.
Alita: Self-Evolving Agents with Minimal Predefinition and Maximal Self-Evolution
Alita achieves top performance on GAIA benchmark (75.42% on test, outperforming OpenAI Deep Research and Manus) through a radically simple design philosophy: minimal predefinition and maximal self-evolution. By dynamically generating and reusing Model Context Protocols (MCPs) on-the-fly, Alita autonomously creates, refines, and adapts capabilities based on task demands, demonstrating that simplicity is the ultimate sophistication in building generalist AI agents.
Research Areas
Foundation Models
Developing foundation AI models such as LLMs and Diffusion Models, for building more capable, efficient, and controllable generative AI systems.
Agentic AI
Developing autonomous AI agents that can reason, plan, take actions, and collaborate to solve complex real-world problems.
Reinforcement Learning
Advancing reinforcement learning methods for training LLMs to reason, sequential decision-making in complex, uncertain environments with applications to robotics and control.
AI for Science and Engineering
Building autonomous AI co-scientists and self-driving labs that can reason, experiment, and discover novel solutions across diverse domains including materials science, drug discovery, and engineering design.
Machine Learning Theory
Fundamental understanding of machine learning algorithms, including generalization bounds, dimensionality reduction, optimization landscapes, and convergence analysis.
Optimization
Novel optimization algorithms for solving large-scale problems in machine learning, operations research, and engineering applications.