Research map / Continual learning / Class-incremental
More Classifiers, Less Forgetting: A Generic Multi-classifier Paradigm for Incremental Learning
Yu Liu, Sarah Parisot, Gregory Slabaugh, Xu Jia, Ales Leonardis, Tinne Tuytelaars
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 multiple side classifiers (K=3) to MAS parameter regularization | MAS · CIFAR-100 and Tiny ImageNet | accuracy: about 4%~5% gains on CIFAR-100, 3%~4% on Tiny ImageNet | Improved |
| add multiple side classifiers to LwF activation regularization | LwF · CIFAR-100 and Tiny ImageNet | accuracy: 3%~5% gains | Improved |
| increase side classifiers from K=2 to K=3 (ablation) | MUC-MAS · CIFAR-100 and Tiny ImageNet | accuracy: about 1.5% improvement | Improved |
| increase side classifiers from K=3 to K=4 or 5 (ablation) | MUC-MAS · CIFAR-100 and Tiny ImageNet | accuracy | No clear effect |
| add side classifiers (MUC) to MAS and LwF | MAS, LwF · CIFAR-100 and Tiny ImageNet | forgetting ratio: -0.55 -> -0.50 (MAS, CIFAR-100, 10 tasks) | Improved |
| add classifier discrepancy loss to side classifier training (ablation) | MUC · CIFAR-100 | Euclidean distance between side classifier parameters: 0.057 -> 5.284 | Improved |
| train in two stages instead of one stage (ablation) | MUC-MAS · CIFAR-100 (100 classes seen) | accuracy: 28.8 -> 31.7 | Improved |
| train in two stages instead of one stage (ablation) | MUC-LwF · CIFAR-100 (100 classes seen) | accuracy: 38.4 -> 41.6 | Improved |
| train in two stages instead of one stage (ablation) | MUC-MAS · CIFAR-100 (20 classes seen) | accuracy: 83.5 -> 82.6 | Hurt |
| increase lambda from 0.01 to 0.02 (ablation) | MAS · CIFAR-100 (100 classes seen) | accuracy: 27.5 -> 30.8 | Improved |
| add side classifiers at fixed lambda 0.01 | MAS · CIFAR-100 (100 classes seen) | accuracy: 27.5 -> 31.7 | Improved |
| add stability factor to MUC-MAS (ablation) | MUC-MAS · CIFAR-100 (100 classes seen) | accuracy: 30.8 -> 31.7 | Improved |
Closest papers by meaning
- A continual learning survey: Defying forgetting in classification tasks (TPAMI 2021 · similarity 0.91)
- Improving Continual Learning Performance and Efficiency with Auxiliary Classifiers (ICML 2025 · similarity 0.89)
- Multi-layer Rehearsal Feature Augmentation for Class-Incremental Learning (ICML 2024 · similarity 0.88)
- Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning (NeurIPS 2024 · similarity 0.88)
- Continual learning with the neural tangent ensemble (NeurIPS 2024 · similarity 0.87)
- AnaCP: Toward Upper-Bound Continual Learning via Analytic Contrastive Projection (NeurIPS 2025 · similarity 0.87)
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.
