Research map / Continual learning / Online and replay-based
Dual-CBA: Improving Online Continual Learning via Dual Continual Bias Adaptors From a Bi-level Optimization Perspective
Hong Wang, Renzhen Wang, Yichen Wu, Xixi Jia, Minghao Zhou, Deyu Meng
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 class-specific CBA to ER (ablation) | ER · Split CIFAR-100 (M=2k) | ACC (%): 21.66 -> 25.56 | Improved |
| upgrade class-specific CBA to Dual-CBA on ER | ER-CBA · Split CIFAR-100 (M=2k) | ACC (%): 25.56 -> 29.14 | Improved |
| add Dual-CBA to ER | ER · Split CIFAR-10 (M=0.2k) | FM (%): 49.81 -> 28.24 | Improved |
| add Dual-CBA to DER++ | DER++ · Split CIFAR-100 (M=5k) | ACC (%): 17.36 -> 29.87 | Improved |
| add Dual-CBA to RAR | RAR · Split CIFAR-10 (M=0.5k) | ACC (%): 47.66 -> 47.68 | No clear effect |
| add class-agnostic CBA to individual class-specific CBA (ablation) | Dual-CBA · Split CIFAR-100 (M=5k) | ACC (%): 31.63 -> 31.86 | Improved |
| use individual class-specific CBAs for old and new tasks instead of a single one (ablation) | ER-Dual-CBA · Split CIFAR-100 (M=2k) | ACC (%): 28.79 -> 29.31 | Improved |
| add class-specific CBA to class-agnostic CBA (ablation) | ER-Dual-CBA-1 · Split CIFAR-100 (M=2k) | ACC (%): 29.76 -> 29.81 | No clear effect |
| replace BN with Incremental BN (ablation) | ER-Dual-CBA · Split CIFAR-100 | ACC (%): 25.29 -> 29.14 | Improved |
| replace BN with Incremental BN (ablation) | ER · Split CIFAR-100 | FM (%): 33.21 -> 33.16 | No clear effect |
| replace BN with Incremental BN (ablation) | DER++ · Split CIFAR-100 | FM (%): 41.35 -> 42.47 | Hurt |
| add class-specific CBA under semi-supervised label ratio 0.2 | ER · Split CIFAR-100 (M=2k) | ACC (%): 16.38 -> 15.98 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SS-IL — SS-IL: Separated Softmax for Incremental Learning (ICCV 2021) · better on 1 of 1 shared result
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · 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
- Enhancing Online Continual Learning with Plug-and-Play State Space Model and Class-Conditional Mixture of Discretization (CVPR 2025 · similarity 0.88)
- Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement (AAAI 2024 · similarity 0.88)
- Adaptive Retention & Correction: Test-Time Training for Continual Learning (ICLR 2025 · similarity 0.88)
- LoRanPAC: Low-rank Random Features and Pre-trained Models for Bridging Theory and Practice in Continual Learning (ICLR 2025 · similarity 0.88)
- Adaptive Shortcut Debiasing for Online Continual Learning (AAAI 2024 · similarity 0.87)
- Ferret: An Efficient Online Continual Learning Framework under Varying Memory Constraints (CVPR 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.
