mlsanityData to 14 Oct 2026

ICED: Concept-level Machine Unlearning via Interpretable Concept Decomposition

arXiv 2605.14309 ↗First posted 14 May 2026cs.CV cs.AI cs.LGcited by 0

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness

also in 1 more
Abstract (arXiv)

Machine unlearning in Vision-Language Models (VLMs) is typically performed at the image or instance level, making it difficult to precisely remove target knowledge without affecting unrelated semantics. This issue is especially pronounced since a single image often contains multiple entangled concepts, including both target concepts to be forgotten and contextual information that should be preserved. In this paper, we propose an interpretable concept-level unlearning framework for VLMs, which constructs a compact task-specific concept vocabulary from the forgetting set using a multimodal large language model. In addition to modality alignment, visual representations are decomposed into sparse, nonnegative combinations of semantic concepts, providing an explicit interface for fine-grained knowledge manipulation. Based on this decomposition, our method formulates unlearning as concept-level optimization, where target concepts are selectively suppressed while intra-instance non-target semantics and global cross-modal knowledge are preserved. Extensive experiments across both in-domain and out-of-domain forgetting settings demonstrate that our method enables more comprehensive target forgetting, better preserves non-target knowledge within the same image, and maintains competitive model utility compared with existing VLM unlearning methods.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 22 of its 37 references that are in the map. First posted on arXiv 14 May 2026.

Builds on

22 of 37 references.

References · 22

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2019Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networkscited by 618
  3. 2022Post-hoc Concept Bottleneck Modelscited by 293
  4. 2021Unrolling SGD: Understanding Factors Influencing Machine Unlearningcited by 279
  5. 2021Machine Unlearning of Features and Labelscited by 229
  6. 2021Trustworthy AI: From Principles to Practicescited by 207
  7. 2023Interpreting CLIP's Image Representation via Text-Based Decompositioncited by 196
  8. 2022Can Bad Teaching Induce Forgetting? Unlearning in Deep Networks using an Incompetent Teachercited by 194
  9. 2020BREEDS: Benchmarks for Subpopulation Shiftcited by 155
  10. 2023Boundary Unlearningcited by 134
  11. 2022Zero-Shot Machine Unlearningcited by 133
  12. 2024Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)cited by 119
10 more
  1. 2023Text-To-Concept (and Back) via Cross-Model Alignmentcited by 64
  2. 2022Do Vision-Language Pretrained Models Learn Composable Primitive Concepts?cited by 31
  3. 2023STAIR: Learning Sparse Text and Image Representation in Grounded Tokenscited by 23
  4. 2023MultiDelete for Multimodal Machine Unlearningcited by 20
  5. 2024CLIPErase: Efficient Unlearning of Visual-Textual Associations in CLIPcited by 19
  6. 2025Learning to Unlearn while Retaining: Combating Gradient Conflicts in Machine Unlearningcited by 19
  7. 2024Targeted Unlearning with Single Layer Unlearning Gradientcited by 18
  8. 2024Zero-Shot Class Unlearning in CLIP with Synthetic Samplescited by 17
  9. 2024An Information Theoretic Approach to Machine Unlearningcited by 16
  10. 2025Targeted Forgetting of Image Subgroups in CLIP Modelscited by 6

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025SAUCE: Selective Concept Unlearning in Vision-Language Models with Sparse Autoencoders
  2. 2024Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models
  3. 2026Null Space Constrained Contrastive Visual Forgetting for MLLM Unlearning
  4. 2026Stochastic Meta-Unlearning: Bridging Language Backbone and Multimodal Unlearning
  5. 2026Knowledge Vector Weakening: Efficient Training-free Unlearning for Large Vision-Language Models
  6. 2026On the Robustness of Machine Unlearning for Vision-Language Models
  7. 2026Suppression Is Not Forgetting: Residual Recoverability in Visual Concept Unlearning for VLMs
  8. 2026One Modality to Forget Them All: Enhancing Cross-Modal Unlearning in Vision-Language Models