Research map / Continual learning / Online and replay-based
KSM: Fast Multiple Task Adaption via Kernel-Wise Soft Mask Learning
Li Yang, Zhezhi He, Junshan Zhang, Deliang Fan
Check 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 |
|---|---|---|---|
| use KSM instead of CPG | continual learning on Twenty Tasks of CIFAR-100 · Twenty Tasks of CIFAR-100 | average accuracy: 78.7 -> 79.2 | Improved |
| use KSM instead of CPG | continual learning on Twenty Tasks of CIFAR-100 · Twenty Tasks of CIFAR-100 | training time (s): 2046 -> 94.3 | Improved |
| use KSM instead of the strongest reported baseline | ImageNet-to-Sketch continual learning · CUBS | accuracy: 83.59 -> 83.81 | Improved |
| use KSM instead of the strongest reported baseline | ImageNet-to-Sketch continual learning · Cars | accuracy: 92.80 -> 92.14 | Hurt |
| use KSM instead of the strongest reported baseline | ImageNet-to-Sketch continual learning · Flowers | accuracy: 96.6 -> 96.94 | Improved |
| use KSM instead of the strongest reported baseline | ImageNet-to-Sketch continual learning · WikiArt | accuracy: 77.15 -> 75.25 | Hurt |
| use KSM instead of the strongest reported baseline | ImageNet-to-Sketch continual learning · Sketch | accuracy: 80.78 -> 81.12 | Improved |
| add scaling factors to the weights zeroed by the binary mask (ablation) | Piggyback on ImageNet-to-Sketch · ImageNet-to-Sketch (CUBS, Cars, Flowers, WikiArt, Sketch) | accuracy: 81.59 -> 82.26; 89.62 -> 91.17; 94.77 -> 95.85; 71.33 -> 73.12; 79.91 -> 80.22 | Improved |
| use the softmax trick instead of Piggyback's binary-mask learning (ablation) | mask learning on ImageNet-to-Sketch · ImageNet-to-Sketch (CUBS, Cars, Flowers, WikiArt, Sketch) | accuracy: 81.59 -> 82.86; 89.62 -> 91.71; 94.77 -> 96.67; 71.33 -> 74.06; 79.91 -> 80.70 | Improved |
| use kernel-wise rather than element-wise masks (ablation) | Piggyback on ImageNet-to-Sketch · ImageNet-to-Sketch (CUBS, Cars, Flowers, WikiArt, Sketch) | accuracy: 81.59 -> 81.76; 89.62 -> 89.57; 94.77 -> 94.88; 71.33 -> 70.30; 79.91 -> 79.95 | Mixed |
| use element-wise rather than kernel-wise masks (ablation) | full KSM on ImageNet-to-Sketch · ImageNet-to-Sketch (CUBS, Cars, Flowers, WikiArt, Sketch) | accuracy: 83.81 -> 83.79; 92.14 -> 92.18; 96.94 -> 96.90; 75.25 -> 75.0; 81.12 -> 81.10 | Mixed |
| vary the initial task (ablation) | KSM on Twenty Tasks of CIFAR-100 · Twenty Tasks of CIFAR-100 | average accuracy: 79.2; 78.8; 79.3; 78.0; 78.2 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Piggyback — Piggyback: Adapting a Single Network to Multiple Tasks by Learning to Mask Weights (ECCV 2018) · better on 0 of 1 shared result
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
- Parameter-Level Soft-Masking for Continual Learning (ICML 2023 · similarity 0.87)
- Piggyback: Adapting a Single Network to Multiple Tasks by Learning to Mask Weights (ECCV 2018 · similarity 0.86)
- Probabilistic Group Mask Guided Discrete Optimization for Incremental Learning (ICML 2025 · similarity 0.84)
- Continual Learning Using a Kernel-Based Method Over Foundation Models (AAAI 2025 · similarity 0.83)
- Gradual Divergence for Seamless Adaptation: A Novel Domain Incremental Learning Method (ICML 2024 · similarity 0.83)
- MOS: Model Surgery for Pre-Trained Model-Based Class-Incremental Learning (AAAI 2025 · similarity 0.83)
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.
