blindness problem
Towards Healing the Blindness of Score Matching
Zhang, Mingtian, Key, Oscar, Hayes, Peter, Barber, David, Paige, Brooks, Briol, François-Xavier
Score-based divergences have been widely used in machine learning and statistics applications. Despite their empirical success, a blindness problem has been observed when using these for multi-modal distributions. In this work, we discuss the blindness problem and propose a new family of divergences that can mitigate the blindness problem. We illustrate our proposed divergence in the context of density estimation and report improved performance compared to traditional approaches.