ICML 2026 Recordings Feature Motion Attribution Talk

Xindi Wu posted on X that the ICML 2026 oral recordings are now live. She directs attention to her talk on motion attribution for video generation. The presentation examines how different training examples influence generated motion and what this reveals about video generation models. Wu is identified in the post as a Princeton PhD student in computer science who focuses on data-centric multimodal machine learning and has interned at NVIDIA AI.