Null Space Constrained Contrastive Visual Forgetting for MLLM Unlearning
📰 ArXiv cs.AI
Learn to unlearn target visual knowledge in MLLMs while preserving non-target knowledge using null space constrained contrastive visual forgetting
Action Steps
- Implement null space constrained contrastive visual forgetting in your MLLM architecture to forget target visual knowledge
- Use contrastive learning to distinguish between target and non-target visual knowledge
- Apply constraints to the null space of the visual embedding to preserve non-target knowledge
- Evaluate the effectiveness of the unlearning approach using metrics such as forgetting rate and retention rate
- Fine-tune the MLLM model after unlearning to ensure minimal impact on non-target knowledge
Who Needs to Know This
Researchers and engineers working on multimodal large language models (MLLMs) can benefit from this approach to selectively forget visual knowledge while retaining other knowledge
Key Insight
💡 Null space constrained contrastive visual forgetting can effectively remove target visual knowledge from MLLMs while preserving non-target knowledge
Share This
🚫 Forget unwanted visual knowledge in MLLMs with null space constrained contrastive visual forgetting 📸
Key Takeaways
Learn to unlearn target visual knowledge in MLLMs while preserving non-target knowledge using null space constrained contrastive visual forgetting
Full Article
Title: Null Space Constrained Contrastive Visual Forgetting for MLLM Unlearning
Abstract:
arXiv:2605.05909v1 Announce Type: new Abstract: The core challenge of machine unlearning is to strike a balance between target knowledge removal and non-target knowledge retention. In the context of Multimodal Large Language Models (MLLMs), this challenge becomes even more pronounced, as knowledge is further divided into visual and textual modalities that are tightly intertwined. In this paper, we introduce an MLLM unlearning approach that aims to forget target visual knowledge while preserving
Abstract:
arXiv:2605.05909v1 Announce Type: new Abstract: The core challenge of machine unlearning is to strike a balance between target knowledge removal and non-target knowledge retention. In the context of Multimodal Large Language Models (MLLMs), this challenge becomes even more pronounced, as knowledge is further divided into visual and textual modalities that are tightly intertwined. In this paper, we introduce an MLLM unlearning approach that aims to forget target visual knowledge while preserving
DeepCamp AI