Zihao He

Flow-Map Distillation achieves superior image restoration by transforming static knowledge transfer into a dynamic flow mapping process, cutting training variance in half.

Asynchronous multimodal policies can significantly outperform traditional synchronous methods by leveraging native inference rates and dynamic guidance.

MLLMs can now judge more consistently and generalize better thanks to a multi-task reinforcement learning approach that aligns them with human preferences across diverse visual tasks.