Wen Yan

Actively repelling uncertain features during representation learning slashes extreme segmentation errors in ambiguous anatomical boundaries by 22x, even without sacrificing overall accuracy.

Amortizing quantum physical constraints into a one-step flow model via surrogate RL nearly doubles transition-state recovery on unseen reactions while running 30× faster than guided diffusion baselines.

Suppressing certain clinical covariates can actually improve the predictive power of prostate MRI grading models, revealing hidden dependencies that could mislead performance assessments.