Decoupled Entropy Minimization
Published in Advances in Neural Information Processing Systems (NeurIPS), 2026
This work proposes decoupled entropy minimization, a novel approach that separates the entropy minimization objective into meaningful components. By decoupling different sources of predictive uncertainty, the method achieves more stable and effective adaptation across distribution shifts.
Recommended citation: Jing Ma, Hanlin Li, Xiang Xiang. (2026). "Decoupled Entropy Minimization." NeurIPS 2026, 38: 161195–161230.
Download Paper