I’m a second-year master’s student at the Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University, supervised by Prof. Xiaolin Huang.

My research focuses on efficient adaptation for large language models (LLMs) and vision-language models (VLMs). Lately, I’ve been exploring scalable MoE-based fine-tuning for VLMs. And I’m excited about agentic LLMs and tool-augmented systems as a direction for building more capable and reliable models.

📝 Featured Publications

ICLR 2026

Bi-LoRA: Efficient Sharpness-Aware Minimization for Fine-Tuning Large-Scale Models

Yuhang Liu^*, Tao Li^*, Zhehao Huang, Zuopeng Yang, Xiaolin Huang

Code

This paper identifies a key gap in parameter-efficient fine-tuning: LoRA is memory-efficient but can generalize poorly, while a naive LoRA+SAM incurs two-step overhead with perturbations confined to a low-rank subspace, and introduces Bi-LoRA, a dual-adapter design that decouples perturbation modeling from task optimization to enable single-backward sharpness-aware training via bi-directional updates with no extra inference cost.

📖 Educations

• 2024.09 - 2027.03 (expected), M.Eng., Electronic and Information Engineering, Shanghai Jiao Tong University, China
• 2020.09 - 2024.06, B.Eng., Automation (IEEE honor class), Shanghai Jiao Tong University, China

💻 Experience

• 2025.08 - present, Intern Algorithm Engineer, Yinwang, Research on improving long-tail generalization for autonomous driving.
• 2024.06 - 2022.05, Intern Algorithm Engineer, HUAWEI, Research on continual post-training for text-to-image diffusion models.