CUSE: Fine-Grained Multimodal Unlearning with Counterfactual Guidance and On-Policy Self-Distillation

Key points

• Multimodal large language models (MLLMs) can reproduce sensitive, copyrighted, or unsafe information from their training data.1
• Effective unlearning must remove target information while preserving non-target content, even within the same response.1
• Existing methods typically apply uniform forgetting objectives to complete responses and rely on fixed, off-policy supervision, resulting in coarse token control and misalignment with the model's current behavior.1
• It first constructs counterfactual inputs by removing target-related evidence and uses the resulting changes in token probabilities to identify which tokens require stronger forgetting.1

Sources (1)

• [1]CUSE: Fine-Grained Multimodal Unlearning with Counterfactual Guidance and On-Policy Self-Distillation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 10, 09:37 AM

  We propose CUSE, a Counterfactual-guided Unlearning framework via on-policy Self-distillation for multimodal large language models.

  Multimodal large language models (MLLMs) can reproduce sensitive, copyrighted, or unsafe information from their training data.

Extractive summary: sentences quoted from the sources.