Introduced the Legal Fact Prediction (LFP) task and constructed a dataset of 657 Chinese civil litigation cases covering 10 representative case types; paper to appear at EMNLP 2025
Launched SAS (Smart Agent Survey), a software using LLM agents to simulate survey responses from uploaded questionnaires (DOCX, PDF, etc.)
Analyzed 7 security challenges in cross-domain multi-agent LLM systems
Studied probabilistic kernel functions for fast angle testing, applied to MIPS and ANNS
Proposed a solution for range-filtering ANNS
Explored novel findings on fusing Vision-Language Models (VLMs) with Large Language Models (LLMs)
Investigated data overvaluation attacks in federated learning and proposed Truth-Shapley as a truthful data valuation metric