UnStar: Unlearning with Self-Taught Anti-Sample Reasoning for LLMs

TL;DR

UnSTAR introduces a novel anti-sample-based unlearning method for LLMs, enabling efficient, targeted removal of specific knowledge without affecting related information, advancing privacy and model control.

Contribution

The paper presents the concept of anti-sample-induced unlearning, a method to generate misleading rationales for targeted unlearning in LLMs, which was not previously explored.

Findings

Anti-samples effectively reverse learned associations.

The method allows fine-grained, targeted unlearning.

Anti-samples accelerate the unlearning process.

Abstract

The key components of machine learning are data samples for training, model for learning patterns, and loss function for optimizing accuracy. Analogously, unlearning can potentially be achieved through anti-data samples (or anti-samples), unlearning method, and reversed loss function. While prior research has explored unlearning methods and reversed loss functions, the potential of anti-samples remains largely untapped. In this paper, we introduce UnSTAR: Unlearning with Self-Taught Anti-Sample Reasoning for large language models (LLMs). Our contributions are threefold; first, we propose a novel concept of anti-sample-induced unlearning; second, we generate anti-samples by leveraging misleading rationales, which help reverse learned associations and accelerate the unlearning process; and third, we enable fine-grained targeted unlearning, allowing for the selective removal of specific associations without impacting related knowledge - something not achievable by previous works. Results demonstrate that anti-samples offer an efficient, targeted unlearning strategy for LLMs, opening new avenues for privacy-preserving machine learning and model modification.