Zheng Jiang

Weak models can supercharge strong models through efficient policy shift transfers, achieving significant performance gains without the usual rollout costs.

Browser agents can achieve unprecedented scalability by harnessing the collective skills of internet users through skill distillation.

Medical VLMs can achieve state-of-the-art visual reasoning without any human annotations, thanks to a novel reinforcement learning framework that leverages model uncertainty and rollout agreement.

SubFLOT tackles federated learning's heterogeneity problem by cleverly using optimal transport to create personalized submodels on the server, sidestepping the computational burden of client-side pruning.