Picture for Zhongyu Wei

Zhongyu Wei

Fudan University

Autoregressive Semantic Visual Reconstruction Helps VLMs Understand Better

Add code
Jun 10, 2025
Viaarxiv icon

AutoJudger: An Agent-Driven Framework for Efficient Benchmarking of MLLMs

Add code
May 27, 2025
Viaarxiv icon

Not All Models Suit Expert Offloading: On Local Routing Consistency of Mixture-of-Expert Models

Add code
May 21, 2025
Viaarxiv icon

OViP: Online Vision-Language Preference Learning

Add code
May 21, 2025
Viaarxiv icon

EcoLANG: Efficient and Effective Agent Communication Language Induction for Social Simulation

Add code
May 11, 2025
Viaarxiv icon

SocioVerse: A World Model for Social Simulation Powered by LLM Agents and A Pool of 10 Million Real-World Users

Add code
Apr 14, 2025
Viaarxiv icon

Word Form Matters: LLMs' Semantic Reconstruction under Typoglycemia

Add code
Mar 03, 2025
Viaarxiv icon

Stepwise Informativeness Search for Improving LLM Reasoning

Add code
Feb 21, 2025
Viaarxiv icon

How Jailbreak Defenses Work and Ensemble? A Mechanistic Investigation

Add code
Feb 20, 2025
Viaarxiv icon

Multi-Agent Simulator Drives Language Models for Legal Intensive Interaction

Add code
Feb 08, 2025
Viaarxiv icon