Private Wasserstein Distance with Random Noises

• DOI: 10.48550/arxiv.2404.06787
• Publication Date: 2024-04-10
• Authors: Wenqian Li; Haozhi Wang; Zhe Huang; Yan Pang
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Machine Learning (cs.LG)

Wasserstein distance is a principle measure of data divergence from distributional standpoint. However, its application becomes challenging in the context privacy, where sharing raw restricted. Prior attempts have employed techniques like Differential Privacy or Federated optimization to approximate distance. Nevertheless, these approaches often lack accuracy and robustness against potential attack. In this study, we investigate underlying triangular properties within space, leading straightforward solution named TriangleWad. This approach enables computation between datasets stored across different entities. Notably, TriangleWad 20 times faster, making information truly invisible, enhancing resilience attacks, without sacrificing estimation accuracy. Through comprehensive experimentation various tasks involving both image text data, demonstrate superior performance generalizations.