Yige Liu works on hidden-state-to-actuation (Hidden to Actuation, HtA), using a model’s internal representations to drive lip, face, and body motion directly so that interaction is more responsive and coherent. Current interests also include test-time training (TTT) and JEPA.
正在做隐空间到执行器相关的工作（Hidden to Actuation，HtA），利用模型内部的表征直接驱动唇部、面部、肢体动作去提升交互实时性以及整体效果，关注 test time training（TTT）、JEPA 等。