Publications
You can also find my articles on my Google Scholar profile.
(* denotes equal contribution.)
2026
PUMA detects when the reasoning of a large reasoning model has semantically converged and exits early, cutting tokens by 26.2% on average while preserving accuracy and chain-of-thought quality.
/ 
MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs ICML 2026
MVI-Bench is the first comprehensive benchmark for evaluating how misleading visual inputs undermine the robustness of LVLMs. Results across 18 state-of-the-art LVLMs reveal pronounced vulnerabilities and provide actionable insights toward more reliable models.
/ 
2025
Enhancing Multimodal In-Context Learning for Image Classification through Coreset Optimization ACM MM 2025, Oral
KeCO is a coreset construction framework for image classification that strengthens the in-context learning of LVLMs. It aggregates category-relevant information from untapped support-set data via feature-level updates, and remains strong in a simulated online setting.
2024
We study how to configure effective in-context sequences for VQA to enhance the in-context learning ability of LVLMs, analyze the role of different configurations, and design new configuration methods that improve ICL performance.
