Research

An Agent-RAG framework that organizes knowledge as method nodes with dual-tree indexing, strategy-guided synthesis, and verification-based pruning, enabling explainable and controllable innovation.

A group-aware EEG emotion recognition framework that combines individual neural dynamics, learnable group prototypes, and multi-subject resonance modeling to improve cross-subject generalization.

A layered multi-agent workflow for repository-level bug and vulnerability detection: risk triage → context expansion → specialist agents → selective verification → evidence fusion, aiming for higher recall, fewer false positives, and better interpretability.