Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception
Learning fine-grained visual detail through region-to-image distillation.
Lai Wei 魏来
PhD student at Shanghai Jiao Tong University
I currently work on scalable pre-training and efficient post-training to advance the perceptual and reasoning capabilities of multimodal LLMs.
I am advised by Weiran Huang at SJTU and jointly trained at Zhongguancun Academy with Chenglin Liu. I currently intern with the Base Model Group at Zhipu AI.
Research
Selected publications.
CLBench-V: Evaluating Multimodal Context Learning from Grounding to Knowledge Acquisition
Evaluating how models ground, apply, and learn new knowledge from multimodal context.
Attend to Evidence: Evidence-Anchored Spatial Attention Supervision for Multimodal RLVR
Guiding reinforcement learning with attention supervision tied to visual evidence.
The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images
A causal audit of whether visual tool observations actually influence model answers.
Targeted Exploration via Unified Entropy Control for Reinforcement Learning
Controlling entropy to guide targeted exploration in reinforcement learning.
First SFT, Second RL, Third UPT: Continual Improving Multi-Modal LLM Reasoning via Unsupervised Post-Training
Unsupervised post-training for continual improvement in multimodal reasoning.
MM-LIMA: Less Is More for Alignment in Multi-Modal Datasets
Selecting small, high-quality instruction datasets for multimodal alignment.
Diabetica: Adapting Large Language Model to Enhance Multiple Medical Tasks in Diabetes Care and Management
Adapting language models to tasks in diabetes care and management.
Diff-eRank: A Novel Rank-Based Metric for Evaluating Large Language Models
A rank-based perspective on evaluating large language models.
Background
Along the way.
Research experience
2026 — present
Zhipu AI
Research intern · Base Model Group
Mentor: Zhengxiao Du
2025 — 2026
Ant Group
Research intern · Machine Intelligence
Mentor: Jun Lan
2024
Tencent
Research intern · WeChat Group
Mentor: Guilin Li
Education
2024 — present
Shanghai Jiao Tong University
PhD in Computer Science
Advisor: Weiran Huang
2024 — present
Zhongguancun Academy
Joint PhD program
Advisor: Chenglin Liu
2020 — 2024
Shanghai Jiao Tong University
Bachelor of Physics
School of Physics and Astronomy
Recognition
Honors & awards.
- 2026Top 3%
National Scholarship
Shanghai Jiao Tong University
- 2025Top 5%
CAST PhD Special Program
中国科协青年科技人才培育工程博士生专项计划
- 2025Top 1.5%
Zhongguancun Academy Scholarship
- 2025
Merit Student
Shanghai Jiao Tong University
Earlier awards 2021–2024
- 2024Top 10%
Outstanding Graduate
Shanghai Jiao Tong University
- 2022Top 10%
Huawei Scholarship
Shanghai Jiao Tong University
- 2021, 2022
Undergraduate B Scholarship
Shanghai Jiao Tong University
- 2021
First Prize · China Undergraduate Mathematical Contest in Modeling
Shanghai Division
- 2021
First Prize Scholarship
School of Physics and Astronomy, Shanghai Jiao Tong University
Community
Academic service.
- Reviewing
- ICLR (2025, 2026), NeurIPS (2025), ICML (2026), CVPR (2026)
- Teaching
- Teaching assistant, Machine Learning at SJTU
Sep 2024 — Jun 2025
Beyond research
Off the page.
I enjoy basketball, football, and badminton. I was captain of my college basketball team and a member of the football team.
Music is another constant: electronic organ (Grade 8) and piano. I also enjoy 2K, FIFA, and Game for Peace.