Research map / Continual learning / Class-incremental
SS-IL: Separated Softmax for Incremental Learning
Hongjoon Ahn, Jihwan Kwak, Subin Lim, Hyeonsu Bang, Hyojun Kim, Taesup Moon
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 |
|---|---|---|---|
| raise replay mini-batch size from 16 to 64 (ablation) | SS-IL · ImageNet-1K T=10 | average top-1 accuracy: 54.3 -> 68.2 | Improved |
| raise replay mini-batch size from 16 to 64 (ablation) | SS-IL · ImageNet-1K T=20 | average top-1 accuracy: 58.8 -> 58.9 | No clear effect |
| raise replay mini-batch size from 16 to 64 (ablation) | SS-IL · ImageNet-1K T=5 | average top-1 accuracy: 68.4 -> 68.2 | No clear effect |
| raise replay mini-batch size from 16 to 64 (ablation) | SS-IL · ImageNet-1K T=10 | average top-5 accuracy: 86.6 -> 86.0 | No clear effect |
| add ER mini-batch replay (ablation) | SS-IL · ImageNet-1K T=10, 10k memory | balance of predictions / final-task accuracy | Improved |
| use growing memory (20 per class) instead of fixed 20k memory | SS-IL · ImageNet-1K T=10 | top-1 accuracy drop vs fixed memory: SS-IL 1.3% drop vs EEIL 8.9% drop | Improved |
| SS-IL with 5 samples per class vs baselines with more | SS-IL · ImageNet-1K T=10, base, growing memory | average top-1 accuracy: SS-IL 62.3 vs PODNet 57.3 at 10 per class | Improved |
| large base setting vs base setting for SS-IL | SS-IL · ImageNet-1K T=10, 10 per class | average top-1 accuracy: 63.4 -> 60.5 | Mixed |
Later papers that compared with this method
4 later accepted papers report SS-IL as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Dual-CBA: Improving Online Continual Learning via Dual Continual Bias Adaptors From a Bi-level Optimization Perspective (TPAMI 2026 · better on 1 of 1 shared result)
- Dealing With Cross-Task Class Discrimination in Online Continual Learning (CVPR 2023 · better on 2 of 2 shared results)
- Heterogeneous Forgetting Compensation for Class-Incremental Learning (ICCV 2023 · better on 2 of 2 shared results)
- On the Stability-Plasticity Dilemma of Class-Incremental Learning (CVPR 2023 · better on 1 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
- Boosting Domain Incremental Learning: Selecting the Optimal Parameters is All You Need (CVPR 2025 · similarity 0.84)
- MultiMax: Sparse and Multi-Modal Attention Learning (ICML 2024 · similarity 0.82)
- Attraction Diminishing and Distributing for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.81)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.80)
- Probabilistic Group Mask Guided Discrete Optimization for Incremental Learning (ICML 2025 · similarity 0.80)
- SD-LoRA: Scalable Decoupled Low-Rank Adaptation for Class Incremental Learning (ICLR 2025 · similarity 0.80)
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.
