Browse publications on Google Scholar (top-right) ↗
Resume (English only)
Academic Achievements
Multiple papers accepted to NeurIPS 2025: ExGra-Med (data-efficient multimodal LLM for healthcare), Token Redundancy in 3D Point Cloud Transformers (90–95% computation reduction via token merging), and Over-Optimization in RLHF for LLM Post-Training
MGPath accepted by Transactions on Machine Learning Research; preliminary version presented at ICLR 2025 Workshop
Work on LLaMA-Adapter’s prompt learning accepted at ICML 2025
Inductive Message Passing Network for Human-in-the-Loop Annotation of Mobile Eye Tracking Data published in Scientific Reports (Nature Portfolio)
PiToMe accepted at NeurIPS 2024, proposing a spectrum-preserving token merging method for Transformers
Paper on molecular conformer aggregation network accepted at ICML 2024
Paper on accelerating transformers accepted as an oral presentation at ICLR 2024
Paper on large-scale medical image pre-training using second-order graph matching accepted at NeurIPS 2023
Proposed LoGra-Med, a novel pre-training algorithm incorporating multi-graph alignment for medical multimodal LLMs
Introduced 'On Zero-Initialized Attention', explaining LLaMA-Adapter’s prompt learning through a Mixture of Experts (MoE) perspective