Research map / Continual learning / Class-incremental
Class-Incremental Learning via Dual Augmentation
Fei Zhu, Zhen Cheng, Xu-yao Zhang, Cheng-lin Liu
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 semanAug (implicit semantic augmentation of old classes) to class-mean replay baseline (ablation) | IL2A baseline · CIFAR-100 10 phases | final accuracy: 34.71 -> 42.09 | Improved |
| add classAug (class augmentation by mixing) to baseline (ablation) | IL2A baseline · CIFAR-100 10 phases | final accuracy: 34.71 -> 41.56 | Improved |
| add classAug on top of semanAug (ablation) | IL2A · CIFAR-100 10 phases | final accuracy: 42.09 -> 45.07 | Improved |
| add semanAug on top of classAug (ablation) | IL2A · CIFAR-100 10 phases | final accuracy: 41.56 -> 45.07 | Improved |
| replace Knowledge Distillation with class-mean replay baseline (ablation) | IL2A · CIFAR-100 10 phases | final accuracy: 6.04 -> 34.71 | Improved |
| add classAug (ablation) | IL2A baseline · CIFAR-100 10 phases | new task accuracy and forgetting | Improved |
| use Mixup instead of classAug as regularizer (ablation) | IL2A · CIFAR-100 final accuracy | final accuracy | Hurt |
| use label smoothing instead of classAug (ablation) | IL2A · CIFAR-100 final accuracy | final accuracy | Hurt |
| use diagonal covariance instead of full covariance in semanAug (ablation) | IL2A · CIFAR-100 | final accuracy | Hurt |
| add classAug to baseline OOD detector | CIFAR-10 ResNet-18 · OOD mean | AUROC: 88.57 -> 94.30 | Improved |
| use Mixup instead of classAug | CIFAR-10 ResNet-18 · OOD mean | AUPR-In: 83.35 -> 79.21 (AUPR-In); 88.57 -> 89.90 (AUROC) | Mixed |
Closest papers by meaning
- PASS++: A Dual Bias Reduction Framework for Non-Exemplar Class-Incremental Learning (TPAMI 2025 · similarity 0.91)
- Cross-Class Feature Augmentation for Class Incremental Learning (AAAI 2024 · similarity 0.89)
- Bring Evanescent Representations to Life in Lifelong Class Incremental Learning (CVPR 2022 · similarity 0.89)
- Class-Incremental Learning by Knowledge Distillation With Adaptive Feature Consolidation (CVPR 2022 · similarity 0.89)
- Multi-layer Rehearsal Feature Augmentation for Class-Incremental Learning (ICML 2024 · similarity 0.87)
- eTag: Class-Incremental Learning via Embedding Distillation and Task-Oriented Generation (AAAI 2024 · 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.
