CV

Doctor of Philosophy

University of California, Davis

Computer Science

Supervised by Muhao Chen

Master of Science

Fudan University

Computer Science

Supervised by Yanghua Xiao

Bachelor of Science

Fudan University

Computer Science

Supervised by Yanghua Xiao

Intern

Google Deepmind

Supervised by Wenxuan Zhou

Research Intern

LUKA Group

Supervised by Muhao Chen

Research Intern

OSU NLP Group

Supervised by Yu Su

Research Intern

Meituan

Supervised by Rui Xie

Graduate Fellowship

Video Models Can Reason with Verifiable Rewards

NeurIPS 2026

We propose VideoRLVR, a systematic RLVR recipe that turns video models from visual imitators to visual reasoners.

When Vision Speaks for Sound

arxiv

We characterize models relying on visual cues and ignoring acoustic information as an audio-visual Clever Hans effect, in which models appear audio-grounded, but actually exploit visual-acoustic correlations without verifying whether the audio and visual streams are truly aligned.

Learning Adaptive Reasoning Paths for Efficient Visual Reasoning

arxiv

We propose AVR, an adaptive visual reasoning framework that decomposes visual reasoning into three cognitive functions: visual perception, logical reasoning, and answer application, enabling models to dynamically choose among different response formats.

OmniGuard: Unified Omni-Modal Guardrails with Deliberate Reasoning

arxiv

We propose OmniGuard, a unified framework that leverages deliberate reasoning to establish omni-modal guardrails for LLMs, ensuring safe and appropriate responses across diverse modalities.

Be My Eyes: Extending Large Language Models to New Modalities Through Multi-Agent Collaboration

arxiv

We propose Be My Eyes, a multi-agent collaboration framework that extends the capabilities of LLMs to new modalities through the cooperation of multiple agents.

Is Extending Modality The Right Path Towards Omni-Modality?

ICLR 2026 Workshop on Multimodal Intelligence

This paper the effect of extending modality, the dominant technique for training multimodal models, where an off-the-shelf language model is fine-tuned on target-domain and language data.

Can LLMs Learn to Map the World from Local Descriptions?

ACL 2026

This study investigates whether LLMs, grounded in locally relative human observations, can construct coherent global spatial cognition by integrating fragmented relational descriptions.

MciteBench: A Benchmark for Multimodal Citation Text Generation in MLLMs

EMNLP 2025

We introduce MCiteBench, the first benchmark designed to evaluate and analyze the multimodal citation text generation ability of MLLMs. Our benchmark comprises data derived from academic papers and review-rebuttal interactions, featuring diverse information sources and multimodal content.

Unraveling Cross-Modality Knowledge Conflicts in Large Vision-Language Models

arxiv

We formally define the problem of cross-modality parametric knowledge conflicts by systematically detecing, interpreting, and mitigating them.

From Persona to Personalization: A Survey on Role-Playing Language Agents

TMLR

A survey on Role-Playing Language Agents.

How Easily do Irrelevant Inputs Skew the Responses of Large Language Models?

COLM 2024

We present a comprehensive investigation into the robustness of LLMs to different types of irrelevant information under various conditions.

TravelPlanner: A Benchmark for Real-World Planning with Language Agents

ICML 2024 Spotlight

We propose TravelPlanner, a new planning benchmark that focuses on travel planning, a common real-world planning scenario. It provides a rich sandbox environment, various tools for accessing nearly four million data records, and 1,225 meticulously curated planning intents and reference plans.

Deductive Beam Search: Decoding Deducible Rationale for Chain-of-Thought Reasoning

COLM 2024

Deductive Beam Search, constrained by the principle of deductive reasoning, integrates CoT with step-wise beam search to guide reasoning towards a deducible path.

Towards Visual Taxonomy Expansion

ACMMM 2023

We propose Visual Taxonomy Expansion (VTE), introducing visual features into the taxonomy expansion task.

SLR:A Million-Scale Comprehensive Crossword Dataset for Simultaneous Learning and Reasoning

Neurocomputing

We propose a novel crossword-based NLU task that imparts knowledge information to a model by solving crossword clues and simultaneously trains the model to infer new knowledge from existing knowledge.

End-to-End Entity Linking with Hierarchical Reinforcement Learning

AAAI 2023

We propose to model the Entity Linking task as a hierarchical decision-making process and design a hierarchical reinforcement learning algorithm to solve the problem.

Early Experience Scaling

This project contains open-source code and data reproductions for "Agent Learning via Early Experience", with scaled environments and trajectories of more than 500k trajectories and interaction data. The results also show consistent improvements across different domains.