Fangming Liu

Learned verification strategies, not LLM scale, unlock reliable GUI agent rewards, outperforming GPT-5.4 by 5.36 points with drastically fewer verification tokens.

State-of-the-art image-to-video models are alarmingly vulnerable to visual prompt attacks, achieving up to 100% success in triggering harmful outputs.

Advanced RS MLLMs struggle with negation, but a novel learning method can dramatically enhance their understanding using minimal unlabeled data.

PULSE slashes communication overhead by 89% and boosts training throughput by up to 2.3x, revolutionizing how we scale diffusion models across GPU clusters.

PACT reduces safety violations by over 30% while simultaneously boosting task success, reshaping the landscape of safe robotic manipulation.