Curriculum Vitae

Beijing University of Technology (BJUT)

• GPA: 4.00/4.00 | 95.72/100 | Rank: 1st/60
• Honors: Fan-Gongxiu Honors College (Top 40 selected from 3000+ students)
• Major Courses: Advanced Mathematics (100), Probability and Statistics (98), Data Structures & Algorithms (98), High-level Programming Language (97), Set Theory & Graph Theory (99), Pattern Recognition (97), Computer Organization (97)

North Carolina State University (NCSU)

• Developed AttentiveCSI: a WiFi-based Channel State Information Gesture Recognition with Attention-Enhanced CNN-LSTM Network
• Awarded First Place in Poster Symposium (1/50 participants)

[1] BEACON: Budget-Efficient Discovery of Policy Violations in LLMs via Cognitive-Guided Monte Carlo Tree Search

[2] EviGuard: Evidence-Guided Connector Intervention for Cross-Modal Safety Unlearning in Multimodal LLMs

[3] VidTouch: A Multimodal Benchmark Dataset for Dynamic Visuo-Tactile Fabric Recognition

[4] A Lane Change Decision-Making Method for Intelligent Vehicles in Foggy Days Based on Dynamic Game Theory

Cross-Modal Safety Unlearning in Multimodal LLMs (EviGuard)

• Identified that cross-modal risk in multimodal LLMs concentrates in a low-rank subspace at the visual-language connector, motivating targeted intervention rather than uniform suppression
• Reduced over-refusal rate (SARR) from 30.3% to 22.3%; on OOD benchmark SIUO, achieved 9.8% ASR and 68.0% Safe & Effective rate vs. best baseline (81.2% / 8.0%)
• Contributions: Experimental framework design, algorithm implementation, ablation analysis, manuscript writing

Budget-Constrained Safety Evaluation of LLMs (BEACON)

• Reframed LLM safety evaluation as budget-constrained failure discovery; formalized efficiency-oriented metrics (k-FDQ, NDA, CCR, DV) capturing discovery timing and harm category diversity beyond traditional ASR
• Built Cognitive-Guided MCTS with defense persona profiling and diversity-aware selection; achieved 85.5–100% ASR across 6 frontier LLMs, discovering failures 3.7× faster than strongest baseline (k-FDQ 26 vs. 95)
• Contributions: Problem formulation, attack design, experimental analysis, manuscript preparation

Vision-based Tactile Perception via Multimodal Learning (VidTouch)

• Built VidTouch (145 fabric categories, 440 RGB images / 440 tactile videos), the first dynamic multimodal benchmark for fabric recognition; supports multi-label classification, cross-modal retrieval, and zero-shot generalization
• Designed X3D + EfficientNet + MLP fusion pipeline achieving 98.6% in-category accuracy; revealed zero-shot generalization bottleneck (41.3%), positioning VidTouch as an open challenge benchmark
• Contributions: Dataset construction, model training and optimization, and experimental validation

• Fan-Gongxiu Honors Scholarship (November 2025)
• Xiaomi Entrepreneur Scholarship (November 2025)
• Study Excellence Scholarship (September 2024 & 2025)
• Outstanding Student Honor (April 2024)
• National 2nd Prize - China Robot Competition and Artificial Intelligence Contest (August 2025)