Research map / Continual learning / Online and replay-based
Online Continual Learning with Dynamic Label Hierarchies
Xinrui Wang, Bartłomiej Twardowski, Alexandra Gomez-Villa, Shao-Yuan Li, Songcan Chen
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 HPR to PredLA (ablation) | HALO · CIFAR100 | AAUC: 58.4 -> 59.2 | Improved |
| add HPR to PredLA (ablation) | HALO · iNaturalist | FFAcc: 35.2 -> 40.1 | Improved |
| add HPR to PredLA (ablation) | HALO · FGVC-Aircraft | FFAcc: 24.9 -> 28.9 | Improved |
| add HPR to PredLA (ablation) | HALO · CUB-200 | MS (lower is better): 0.97 -> 0.93 | Improved |
| add PredLA aggregation of linear and analytic heads instead of analytic head alone (ablation) | HALO · CUB-200 | AAUC: 52.5 -> 61.2 | Improved |
| add PredLA aggregation instead of analytic head alone (ablation) | HALO · FGVC-Aircraft | FFAcc: 30.5 -> 24.9 | Hurt |
| add 40% erroneous (noise) hierarchy connections | HALO · iNaturalist | AAUC: 78.06 -> 73.05 | Hurt |
| delete 40% of hierarchy connections (vacancy) | HALO · iNaturalist | AAUC: 78.06 -> 75.88 | Hurt |
| increase replay buffer from 100 to 5000 | HALO · CIFAR100 | AAUC: 52.9 -> 62.5 | Improved |
| increase number of split groups from 5 to 50 | HALO · CIFAR100 | AAUC: 61.2 -> 55.1 | Hurt |
| increase number of split groups from 5 to 50 | HALO · iNaturalist | FAUC: 47.2 -> 53.7 | Improved |
| use HALO instead of ACIL | ACIL · iNaturalist (ViTB / DINO) | AAUC: 80.4 -> 85.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ACIL — ACIL: Analytic Class-Incremental Learning with Absolute Memorization and Privacy Protection (NeurIPS 2022) · better on 8 of 8 shared results
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
- Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning (NeurIPS 2024 · similarity 0.87)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.87)
- Improving Plasticity in Online Continual Learning via Collaborative Learning (CVPR 2024 · similarity 0.87)
- Online Task-Free Continual Learning via Dynamic Expansionable Memory Distribution (CVPR 2025 · similarity 0.87)
- Cut out and Replay: A Simple yet Versatile Strategy for Multi-Label Online Continual Learning (ICML 2025 · similarity 0.86)
- Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation (ICLR 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.
