Research map / Continual learning / Class-incremental
Rebalancing Batch Normalization for Exemplar-Based Class-Incremental Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace BN with TBBN | FT, EEIL, LUCIR, and SSIL · CIFAR-100, 20 tasks | average accuracy: 29.66 -> 34.45; 35.11 -> 39.32; 34.36 -> 37.07; 36.31 -> 38.55 | Helped |
| Replace BN with TBBN | FT, EEIL, LUCIR, and SSIL · ImageNet-100, 20 tasks | average accuracy: 39.06 -> 43.84; 37.89 -> 42.30; 39.34 -> 39.90; 43.84 -> 46.08 | Helped |
| Use CN (G=16) instead of BN | FT, EEIL, LUCIR, and SSIL · CIFAR-100, 20 tasks | average accuracy: 29.66 -> 30.12; 35.11 -> 35.49; 34.36 -> 34.83; 36.31 -> 36.00 | Mixed |
| Replace BN with TBBN | FT, EEIL, LUCIR, and SSIL · CIFAR-100, 6 tasks with a base task | final accuracy: 35.86 -> 37.36; 36.93 -> 38.83; 38.22 -> 39.54; 45.69 -> 46.61 | Helped |
| Replace BN with TBBN | FT · CIFAR-100, online Class-IL, 20 tasks | average accuracy: 10.94 -> 10.12 | Hurt |
| Replace BN with TBBN | FT · CIFAR-100, online Task-IL, 20 tasks | average accuracy: 68.70 -> 67.43 | Hurt |
| Balance batches by augmenting exemplar-memory samples, using ordinary BN | FT · CIFAR-100, 10 tasks | accuracy: 25.54 | Hurt |
| Use a feasible divisor r* instead of the exact computed r when r is not a common divisor (ablation) | TBBN · CIL experiments | accuracy | No clear effect |
Closest papers by meaning
- Gradient Reweighting: Towards Imbalanced Class-Incremental Learning (CVPR 2024 · similarity 0.90)
- Overcoming Recency Bias of Normalization Statistics in Continual Learning: Balance and Adaptation (NeurIPS 2023 · similarity 0.90)
- Learning a Unified Classifier Incrementally via Rebalancing (CVPR 2019 · similarity 0.89)
- Decorrelated Batch Normalization (CVPR 2018 · similarity 0.87)
- Understanding Batch Normalization (NeurIPS 2018 · similarity 0.87)
- Task-recency bias strikes back: Adapting covariances in Exemplar-Free Class Incremental Learning (NeurIPS 2024 · similarity 0.86)
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.