I am currently a PhD student in College of AI, Tsinghua University, supervised by Lu Mi. I am also a member from MAIC team, aiming at next-generation of AI empowered education, leading by Jifan Yu.
My research spans AI, Cognitive science, and AI for education, including human–AI cognitive alignment, agentic research for brain discovery, and human-centric AI systems for education. For a broader scope, I seek to understand the similarities and differences between human and artificial cognition and intelligence, examine how human cognition changes in the age of AI, and use these insights to guide human–AI cognitive collaboration, augmentation, and co-evolution toward human-centric AI.
I also lead BrainPilot, a fully open-source multi-agent platform for trustworthy brain science research, together with its evaluation benchmark, BrainPilotBench.
I am currently a Ph.D. student in the College of AI at Tsinghua University. I received my M.Eng. and B.Eng. degrees from the School of Computer Science and Engineering at Beihang University in 2025 and 2022, respectively. IELTS: 7.5.
arXiv preprint, 2026
BrainPilot is a fully open-source multi-agent system for trustworthy brain science research, with traceable execution, agent-verified results, and the BrainPilotBench-v0 benchmark.
ACL 2026 System Demonstrations
Graph of Trace records fine-grained execution events as real-time visual traces, making scientific-agent workflows easier to inspect and review.
ACL 2026
MASA assesses critical thinking through interpretable, scenario-based multi-agent interaction, bridging the gap between self-reported intentions and observed behavior.
ACL 2026
SimPBL coordinates an orchestrator agent and boundary-aware collaborator agents to provide adaptive scaffolding for project-based learning.
ACL 2026
TeachCraft scaffolds content selection, sequencing, and synthesis through specialized agents to support pedagogical decision-making in lesson generation.
arXiv preprint arXiv:2606.07591, 2026
AI Open, 2026
This work proposes a pipeline to simulate students with learning difficulties using LLMs, supporting metacognitive training through scalable and explainable agents evaluated via a score propagation system.
Frontiers of Digital Education, 2025(Cover Article)
We introduce Explainable Few-shot Knowledge Tracing, a new educational assessment method using large language models to predict student knowledge from limited data with natural language explanations.
arxiv preprint, 2025
We introduce LecEval, a novel metric grounded in multimedia learning theory for automatically evaluating knowledge acquisition in slide-based educational content using four pedagogical rubrics.
JCST, 2026
We propose MAIC, a scalable and adaptive LLM-based multi-agent framework for personalized online education, validated on 100K learning records and designed as a future collaborative open platform.
IEEE Transactions on Computational Scoial Systems, 2024
W-GAT is a novel graph attention network for collabohrative filtering that uses Wasserstein distance to capture user and item uncertainty, improving recommendation accuracy.
Frontiers of Digital Education, 2024
We review data mining advancements in personalized education, covering educational recommendation, cognitive diagnosis, knowledge tracing, and learning analysis, providing structured taxonomies, datasets, and future research directions.
CIKM 2023
We propose PopDCL, a novel contrastive learning framework for collaborative filtering that addresses the popularity bias and sample bias in recommendation systems.
KSEM 2023
We introduce a debiased contrastive loss for collaborative filtering in recommender systems to correct sample bias, enhancing user interest representation. It's integrated into MF and GNN models, improving performance and training efficiency.