Zhiqiang Shen

We propose a data-centric residual matching method that significantly improves dataset distillation efficiency and accuracy.

We propose a simple yet highly effective transfer-based baseline for attacking black-box closed-source LVLMs.

We present Open CaptchaWorld, a benchmark that tests multimodal LLM agents on solving real-world CAPTCHAs via multi-step reasoning and interaction, revealing large gaps between current models and human performance.