Yixin Chen

By pretraining in hyperbolic space, robots can learn more robust spatial perception, leading to significantly improved manipulation performance compared to traditional Euclidean embeddings.

Pathology report generation gets a boost from RANGER, a mixture-of-experts model that dynamically specializes and filters retrieved knowledge, significantly improving generation quality.

Accurately predict Parkinson's surgery outcomes *before* the operation using a new AI model that learns which brain regions matter most for each patient.

Generating realistic tabular data with both numbers and free-form text just got easier: TabDLM bridges the gap between diffusion models and LLMs for superior joint modeling.