Imitation Privacy

TL;DR

This paper introduces the concept of imitation privacy, a new model privacy notion relevant to cloud-based machine learning services, highlighting its differences from traditional data privacy and its broad applicability.

Contribution

The paper develops a general framework for imitation privacy and demonstrates its relevance across various MLaaS scenarios and multi-organizational learning contexts.

Findings

Imitation privacy offers a new perspective on protecting models in ML services.

It applies broadly to query-response MLaaS and multi-organizational learning.

It fundamentally differs from data-level privacy approaches.

Abstract

In recent years, there have been many cloud-based machine learning services, where well-trained models are provided to users on a pay-per-query scheme through a prediction API. The emergence of these services motivates this work, where we will develop a general notion of model privacy named imitation privacy. We show the broad applicability of imitation privacy in classical query-response MLaaS scenarios and new multi-organizational learning scenarios. We also exemplify the fundamental difference between imitation privacy and the usual data-level privacy.