Research map / Image generation / Text-to-image diffusion
Scissorhands: Scrub Data Influence via Connection Sensitivity in Networks
Jing Wu, Mehrtash Harandi
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 gradient projection in the repair phase (ablation) | Scissorhands · CIFAR-10 (forget 10%) | Avg. Gap: 2.92 -> 1.62 | Improved |
| re-initialize trimmed weights with uniform distribution rather than Gaussian or zero (ablation) | Scissorhands · CIFAR-10 (forget 10%) | Avg. Gap: 1.80 (Gaussian), 1.96 (zero) -> 1.62 (uniform) | Improved |
| re-initialize with constant one (ablation) | Scissorhands · CIFAR-10 (forget 10%) | Avg. Gap: 1.62 -> 10.46 | Hurt |
| increase the ratio of forgetting data used for trimming (ablation) | Scissorhands · CIFAR-10 | accuracy metrics | Hurt |
| increase trimming percentage parameter k (fewer weights re-initialized) (ablation) | Scissorhands · CIFAR-10 | accuracy | Improved |
| set balance weight lambda to 0.05 (ablation) | Scissorhands · CIFAR-10 (forget 10%) | test and remaining accuracy | Mixed |
| Scissorhands versus SalUn on CIFAR-100 random-sample forgetting | Scissorhands · CIFAR-100 (forget 10%) | Avg. Gap: 10.45 -> 4.11 | Improved |
| Scissorhands versus SalUn on CelebAMask-HQ identity forgetting | Scissorhands · CelebAMask-HQ (forget 10% identities) | Avg. Gap: 2.93 -> 2.82 | No clear effect |
| Scissorhands versus SalUn on nudity erasure in Stable Diffusion | Scissorhands · Stable Diffusion v1.4 (NudeNet, I2P) | exposed body part counts | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SalUn — SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation (ICLR 2024) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Score Forgetting Distillation: A Swift, Data-Free Method for Machine Unlearning in Diffusion Models (ICLR 2025 · similarity 0.85)
- Targeted Forgetting of Image Subgroups in CLIP Models (CVPR 2025 · similarity 0.85)
- Fast Machine Unlearning without Retraining through Selective Synaptic Dampening (AAAI 2024 · similarity 0.84)
- Machine Unlearning for Image-to-Image Generative Models (ICLR 2024 · similarity 0.84)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.84)
- What makes unlearning hard and what to do about it (NeurIPS 2024 · similarity 0.84)
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.
