Research Overview

Cognitive Intelligence

I study cognitive intelligence: enabling machines to understand, reason, and interact. My work spans three directions with independent scientific questions, connected through autonomous learning, continual evolution, and human–AI collaboration.

Core Research Directions

01 · General Mechanisms of Intelligence

LLM Agents & Autonomous Learning

Study agent harnesses, long-term memory, autonomous interaction, and learning mechanisms that support decision-making, transferable experience, and continual evolution.

Scientific question

How can LLM agents make autonomous decisions, accumulate transferable experience, and continually improve?

Agent HarnessAgent MemorySelf-EvolutionAutonomous Learning

Representative work

Agent training, persistent memory, and evidence-guided research workflows.

Agent-R1 · CIKM 2026 DemoMemWeaver · WWW 2026Mind2Report · CIKM 2026

02 · Research for Scientific Innovation

Scientific Knowledge & Tool Mining

Study scientific literature mining and scientific tool mining, including knowledge extraction and reasoning, tool capability modeling, and autonomous tool orchestration.

Scientific question

How can scientific knowledge and tools be discovered, organized, and reliably used for scientific innovation?

Scientific Literature MiningScientific Tool MiningKnowledge DiscoveryAgentic Research

Representative work

Scientific paper discovery, tool-augmented literature reasoning, and chemical-table understanding.

PaperScout · Findings of EMNLP 2026PaperArena · WSDM 2027ChemTable · KDD 2026

03 · Research for Complex-System Cognition

Cognitive Intelligence for Structured Data

Study semantics, context, relationships, and dynamics in time series, tables, and graphs to support knowledge-grounded reasoning, interactive analysis, and decisions.

Scientific question

How can knowledge and evidence support reasoning about complex relationships and dynamics in structured data?

Time SeriesTabular DataGraph Data

Representative work

Context-aware forecasting and tool-augmented reasoning over time series and tables.

AlphaCast · CIKM 2026Time-R1 · CIKM 2026TableMind · WSDM 2026

How the Directions Connect

LLM agents and autonomous learning provide general mechanisms for cognition. Scientific knowledge and tools, and structured data, pose distinct research questions and provide evidence and feedback that help refine these mechanisms.

Application Domains

Applications include AI for Science through scientific literature and tool mining; Personalized Intelligence through user behavior and contextual memory; and Forecasting Intelligence for electricity load, power generation, and weather, with context-aware prediction and uncertainty estimation.

For research opportunities and collaboration, visit the USTC-AGI Group or contact me by email.