Research map / Label-efficient and robust learning / Recognition and long-tail
Imbalanced Continual Learning with Partitioning Reservoir Sampling
Chris Dongjoo Kim, Jinseo Jeong, Gunhee Kim
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 |
|---|---|---|---|
| use Partitioning Reservoir Sampling instead of conventional reservoir sampling | replay-based multi-label continual learning · COCOseq overall | mAP: 50.2 -> 55.3 | Improved |
| use Partitioning Reservoir Sampling instead of conventional reservoir sampling | replay-based multi-label continual learning · NUS-WIDEseq overall | mAP: 20.1 -> 25.5 | Improved |
| use PRS instead of CRS for minority classes | replay memory · COCOseq minority classes | mAP: 26.9 -> 39.7 | Improved |
| use PRS instead of CRS for majority classes | replay memory · COCOseq majority classes | mAP: 67.9 -> 67.5 | Hurt |
| use PRS instead of CRS with memory size 1000 | replay memory · COCOseq overall | mAP: 47.1 -> 51.2 | Improved |
| use PRS instead of CRS with memory size 3000 | replay memory · COCOseq overall | mAP: 51.1 -> 56.7 | Improved |
| use PRS instead of CRS under task schedule 2,3,0,1 | replay memory · COCOseq overall | mAP: 48.2 -> 53.1 | Improved |
| use PRS instead of CRS under task schedule 3,1,0,2 | replay memory · COCOseq overall | mAP: 49.7 -> 54.1 | Improved |
| move allocation power rho away from the balanced vicinity (ablation) | PRS · COCOseq | C-F1, O-F1, mAP | Hurt |
| add replay with PRS over plain finetuning | online multi-label learner · COCOseq overall | mAP: 16.4 -> 55.3 | Improved |
| use PRS instead of ExStream, which uses task prior information | replay memory · NUS-WIDEseq overall | mAP: 21.0 -> 25.5 | Improved |
Closest papers by meaning
- Online Continual Learning from Imbalanced Data (ICML 2020 · similarity 0.90)
- Towards Macro-AUC Oriented Imbalanced Multi-Label Continual Learning (AAAI 2025 · similarity 0.89)
- Continual Learning on Noisy Data Streams via Self-Purified Replay (ICCV 2021 · similarity 0.87)
- Cut out and Replay: A Simple yet Versatile Strategy for Multi-Label Online Continual Learning (ICML 2025 · similarity 0.87)
- Rebalancing Multi-Label Class-Incremental Learning (AAAI 2025 · similarity 0.86)
- The Inter-batch Diversity of Samples in Experience Replay for Continual Learning (AAAI 2024 · similarity 0.85)
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.
