Protecting Privacy Through Approximating Optimal Parameters for Sequence Unlearning in Language Models

TL;DR

This paper introduces POP, a novel method for unlearning specific data from language models by approximating optimal parameters, effectively balancing privacy protection and model performance.

Contribution

The work proposes a new unlearning technique that approximates optimal gradient updates, improving privacy protection while maintaining model accuracy better than existing methods.

Findings

POP outperforms state-of-the-art unlearning methods across multiple benchmarks.

It effectively forgets target sequences with minimal performance degradation.

Remnant Memorization Accuracy quantifies privacy risks and validates unlearning effectiveness.

Abstract

Although language models (LMs) demonstrate exceptional capabilities on various tasks, they are potentially vulnerable to extraction attacks, which represent a significant privacy risk. To mitigate the privacy concerns of LMs, machine unlearning has emerged as an important research area, which is utilized to induce the LM to selectively forget about some of its training data. While completely retraining the model will guarantee successful unlearning and privacy assurance, it is impractical for LMs, as it would be time-consuming and resource-intensive. Prior works efficiently unlearn the target token sequences, but upon subsequent iterations, the LM displays significant degradation in performance. In this work, we propose Privacy Protection via Optimal Parameters (POP), a novel unlearning method that effectively forgets the target token sequences from the pretrained LM by applying optimal gradient updates to the parameters. Inspired by the gradient derivation of complete retraining, we approximate the optimal training objective that successfully unlearns the target sequence while retaining the knowledge from the rest of the training data. Experimental results demonstrate that POP exhibits remarkable retention performance post-unlearning across 9 classification and 4 dialogue benchmarks, outperforming the state-of-the-art by a large margin. Furthermore, we introduce Remnant Memorization Accuracy that quantifies privacy risks based on token likelihood and validate its effectiveness through both qualitative and quantitative analyses.