Hongyuan Zhang

Treating distillation targets as dynamic on-policy decisions rather than static teacher outputs prevents catastrophic error propagation when adapting compact vision-language models to out-of-distribution multimodal data.

Generative models can significantly boost representation learning, but the reverse is equally transformative for generative quality.

Miles achieves state-of-the-art performance in Class Incremental Learning by efficiently expanding the parameter space without sacrificing computational efficiency or risking catastrophic forgetting.

FrameONE revolutionizes echocardiographic keyframe detection by achieving state-of-the-art accuracy through a unified multi-view approach that overcomes traditional limitations.

Cleaner visual cues can boost multimodal reasoning performance by over 6 points, challenging the notion that simply extending reasoning traces is sufficient.

Imperceptible frequency-aware perturbations can effectively thwart model stealing in text-to-image generative models without sacrificing visual quality.

Forget complex per-sample loss calculations – this simple three-line code injection uses batch loss smoothing to prune 20-50% of training data without sacrificing performance.