My research centers on LLM-driven reasoning and AI agents, with a focus on context-aware reasoning, autonomous interactive, and continual learning and adaptation. This work is motivated by complex tasks in time-series intelligence and science intelligence (scientific knowledge and tool mining). These application-driven directions are framed by 科言 SciToken — understanding the scientific world, and 科语 SciTime — modeling the dynamic world.
How can agents reliably solve problems through reasoning and interaction when information is incomplete, environments change, and feedback is costly?
Developing autonomous interactive learning for large language models, including environment-interactive Agentic RL, tool-augmented reasoning, multi-agent orchestration, and continual capability evolution through context, knowledge, and memory.
科语 SciTime — Modeling the dynamic world. Developing context-aware predictive intelligence, with a focus on multimodal context representation, slow-thinking reasoning, uncertainty-aware forecasting, and autonomous agentic interaction.
科言 SciToken — Understanding the scientific world. Focusing on scientific data, scientific tools, scientific knowledge, and capability enhancement of scientific foundation models.
Real-world settings for developing and evaluating intelligent systems.
Connecting scientific data and knowledge to support reasoning and autonomous discovery.
Forecasting and decision support for complex, evolving real-world systems.
Understanding behaviors and preferences to deliver adaptive, personalized recommendations.