Machine Unlearning in Forgettability Sequence

TL;DR

This paper investigates the varying difficulty of forgetting data samples in machine unlearning, identifies key factors influencing unlearning, and proposes a new framework called RSU to improve unlearning performance.

Contribution

It introduces the concept that sample difficulty varies based on privacy risk and presents the RSU framework with ranking and sequence modules for better unlearning.

Findings

Samples with higher privacy risks are more likely to be forgotten.

Sample difficulty impacts unlearning algorithm performance.

The proposed RSU framework improves unlearning effectiveness.

Abstract

Machine unlearning (MU) is becoming a promising paradigm to achieve the "right to be forgotten", where the training trace of any chosen data points could be eliminated, while maintaining the model utility on general testing samples after unlearning. With the advancement of forgetting research, many fundamental open questions remain unanswered: do different samples exhibit varying levels of difficulty in being forgotten? Further, does the sequence in which samples are forgotten, determined by their respective difficulty levels, influence the performance of forgetting algorithms? In this paper, we identify key factor affecting unlearning difficulty and the performance of unlearning algorithms. We find that samples with higher privacy risks are more likely to be unlearning, indicating that the unlearning difficulty varies among different samples which motives a more precise unlearning mode. Built upon this insight, we propose a general unlearning framework, dubbed RSU, which consists of Ranking module and SeqUnlearn module.