Research map / Vision-language and multimodal / Multimodal LLMs
MultiDelete for Multimodal Machine Unlearning
Jiali Cheng, Hadi Amiri
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add modality decoupling loss (ablation) | MultiDelete · NLVR2 (ALBEF) | deletion set accuracy (Df): 50.3 -> 23.5 | Improved |
| add unimodal knowledge retention loss (ablation) | MultiDelete · NLVR2 (ALBEF) | test accuracy (DTest): 79.2 -> 80.8 | Improved |
| add multimodal knowledge retention loss (ablation) | MultiDelete · NLVR2 (ALBEF) | test accuracy (DTest): 77.1 -> 80.8 | Improved |
| add unimodal knowledge retention loss (ablation) | MultiDelete · Unimodal image classifier accuracy | accuracy: 77.9 -> 83.6 | Improved |
| multimodal unlearning with MultiDelete instead of FineTune | ALBEF · NLVR2 deletion set | accuracy (Df): 49.8 -> 23.5 | Improved |
| weight scrubbing (DtD, L-codec) on all parameters | multimodal models · Flickr30K-IR (ALBEF) | test recall (DTest): 97.8 (Retrain) -> 10.3 (DtD) | Hurt |
| update only the fusion module instead of all parameters (ablation) | MultiDelete · average over tasks (ALBEF) | test performance: 84.2 -> 83.7 | No clear effect |
| update only the fusion module for scrubbing methods (ablation) | DtD · average over tasks (ALBEF) | test performance: 33.0 -> 39.2 | Improved |
| update only the fusion module for modality-agnostic approaches (ablation) | FineTune, NegGrad, UL · average over tasks (ALBEF) | test performance: marginal gap of less than 0.6 absolute points | No clear effect |
| use MultiDelete instead of L-codec for membership inference protection | ALBEF · average MI ratio | prior-to-post MI ratio: 1.19 -> 1.27 | Mixed |
Closest papers by meaning
- Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models (NeurIPS 2024 · similarity 0.89)
- SALMUBench: A Benchmark for Sensitive Association-Level Multimodal Unlearning (CVPR 2026 · similarity 0.89)
- Machine Unlearning for Image-to-Image Generative Models (ICLR 2024 · similarity 0.88)
- LoTUS: Large-Scale Machine Unlearning with a Taste of Uncertainty (CVPR 2025 · similarity 0.86)
- Cross-Modal Unlearning via Influential Neuron Path Editing in Multimodal Large Language Models (AAAI 2026 · similarity 0.86)
- The Illusion of Unlearning: The Unstable Nature of Machine Unlearning in Text-to-Image Diffusion Models (CVPR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
