Hang Li

Treating negative samples based on their similarity to positives leads to a 13.1% boost in retrieval performance, revealing the critical role of grain-level information.

Achieving lossless processing of 256K contexts, Keye-VL-2.0 transforms how we approach long-video understanding and agentic intelligence.

Long-context LLMs can drastically reduce the number of model calls needed for passage re-ranking, achieving efficiency without sacrificing effectiveness.

Forget everything you thought you knew about multimodal agent memory: TaskMem learns what to remember on the fly, boosting VQA accuracy by up to 7% without even looking at the raw video.

Doc-V* demonstrates that an agentic approach to multi-page document VQA, using active navigation and structured memory, can significantly outperform retrieval-augmented generation, especially in out-of-domain scenarios.