Towards Reliable Forgetting: A Survey on Machine Unlearning Verification

TL;DR

This survey addresses the critical challenge of verifying machine unlearning, i.e., confirming that targeted training data has been effectively forgotten. It introduces a two-pronged taxonomy—behavioral verification and parametric verification—grounded in the signals used to assess forgetting, and surveys representative methods within each category. Seven evaluation dimensions are proposed to compare methods on theoretical guarantees, access requirements, granularity, and efficiency, among others. The paper also analyzes reliability threats, including parameter-space forging and behavioral-space deception, and outlines open questions such as unifying the definition of forgetting and developing robust, scalable verification frameworks. Overall, the work provides a structured foundation for advancing verifiable, auditable unlearning in privacy-sensitive and compliance-driven settings.

Abstract

With growing demands for privacy protection, security, and legal compliance (e.g., GDPR), machine unlearning has emerged as a critical technique for ensuring the controllability and regulatory alignment of machine learning models. However, a fundamental challenge in this field lies in effectively verifying whether unlearning operations have been successfully and thoroughly executed. Despite a growing body of work on unlearning techniques, verification methodologies remain comparatively underexplored and often fragmented. Existing approaches lack a unified taxonomy and a systematic framework for evaluation. To bridge this gap, this paper presents the first structured survey of machine unlearning verification methods. We propose a taxonomy that organizes current techniques into two principal categories -- behavioral verification and parametric verification -- based on the type of evidence used to assess unlearning fidelity. We examine representative methods within each category, analyze their underlying assumptions, strengths, and limitations, and identify potential vulnerabilities in practical deployment. In closing, we articulate a set of open problems in current verification research, aiming to provide a foundation for developing more robust, efficient, and theoretically grounded unlearning verification mechanisms.

Figures (4)

• Figure 1: An overview of our work.
• Figure 2: A framework diagram of machine unlearning.
• Figure 3: The process of machine unlearning verification.
• Figure 4: A taxonomy of unlearning verification methods.

Theorems & Definitions (4)

• Definition 1: $\epsilon$-Differential Privacy dp
• Definition 2: ($\epsilon, \delta$)-Differential Privacy dp
• Definition 3: $\epsilon$-Certified Forgetting guo
• Definition 4: $(\epsilon, \delta)$-Certified Forgetting guo