Towards Benign Memory Forgetting for Selective Multimodal Large Language Model Unlearning

TL;DR

This paper introduces SMFA, a novel method for selective unlearning in multimodal large language models that effectively removes sensitive information without degrading overall image understanding, supported by a new comprehensive benchmark.

Contribution

The paper presents SMFA, a targeted unlearning approach that confines forgetting to specific memory regions, and introduces S-MLLMUn Bench, the first benchmark for evaluating sensitive knowledge removal and knowledge retention.

Findings

SMFA achieves precise unlearning of sensitive data.

SMFA maintains the model's general visual understanding.

The new benchmark effectively evaluates unlearning performance.

Abstract

Multimodal Large Language Models (MLLMs) achieve remarkable capabilities but can inadvertently memorize privacy-sensitive information. Although existing unlearning methods can remove such knowledge, they fail to achieve benign forgetting because they often degrade the model's general image understanding performance. To address this, we propose the Sculpted Memory Forgetting Adapter (SMFA), which confines forgetting to targeted memory regions while preserving overall capabilities. SMFA first fine-tunes the model to replace sensitive responses with refusals, yielding a memory forgetting adapter, and then applies a retaining anchor-guided masking mechanism to prevent interference with unrelated knowledge and understanding ability. To systematically evaluate selective MLLM unlearning, we introduce S-MLLMUn Bench, the first benchmark designed to jointly assess the removal of sensitive knowledge and retention of general visual understanding. Extensive experiments show that, unlike prior methods, SMFA achieves precise and controllable unlearning while maintaining the model's foundational image understanding.