Towards Healing the Blindness of Score Matching
Machine Learning
2025-11-25 v3 Machine Learning
Abstract
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.
Keywords
Cite
@article{arxiv.2209.07396,
title = {Towards Healing the Blindness of Score Matching},
author = {Mingtian Zhang and Oscar Key and Peter Hayes and David Barber and Brooks Paige and François-Xavier Briol},
journal= {arXiv preprint arXiv:2209.07396},
year = {2025}
}