Research interests lie in the intersection of AI and Science, including:

Science for AI: Constructing AI models inspired by scientific principles (e.g., KAN, Poisson Flow, Brain-inspired networks). Future interests include deformable neural networks and generative models inspired by physics, among others.

Science of AI: Understanding existing AI models using scientific methods and theories (e.g., understanding Grokking, Neural Scaling Laws, Neural Thermodynamic Laws). Future interests include the science of large models and mechanistic interpretability.

AI for Science: Accelerating scientific discoveries using AI (e.g., AI Poincare for discovering conservation laws). Future interests include identifying major problems in fundamental sciences that could potentially be solved by AI through extensive reading and analysis.