Research map / Continual learning / Online and replay-based
Class-Incremental Continual Learning Into the eXtended DER-Verse
Matteo Boschini, Lorenzo Bonicelli, Pietro Buzzega, Angelo Porrello, Simone Calderara
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 memory logit update (future past) (ablation) | X-DER · Split CIFAR-100 (buffer 500) | FAA: 42.67 -> 49.93 | Improved |
| add memory logit update (future past) (ablation) | X-DER · Split CIFAR-100 (buffer 2000) | FAA: 56.55 -> 59.14 | Improved |
| add memory logit update (future past) (ablation) | X-DER · Split NTU-60 | FAA: 57.66 -> 64.86 | Improved |
| add pre-allocated future heads with contrastive preparation (ablation) | X-DER · Split CIFAR-100 (buffer 500) | FAA: 45.61 -> 49.93 | Improved |
| add pre-allocated future heads with contrastive preparation (ablation) | X-DER · Split miniImageNet (buffer 2000) | FAA: 21.71 -> 28.19 | Improved |
| add pre-allocated future heads with contrastive preparation (ablation) | X-DER · Split NTU-60 | FAF: 9.80 -> 9.95 | Improved |
| use contrastive future preparation instead of fixed RPC heads (ablation) | X-DER · Split CIFAR-100 (buffer 500) | FAA: 48.53 -> 49.93 | Improved |
| use contrastive future preparation instead of CE on future heads (ablation) | X-DER · Split CIFAR-100 (buffer 500) | FAA: 47.67 -> 49.93 | Improved |
| replace DER++ with X-DER | DER++ · Split CIFAR-100 (buffer 500) | FF: 50.54 -> 19.90 | Improved |
| add memory logit update (ablation) | X-DER · CIFAR-100 superclass | SS-ERR: 0.64 -> 0.57 | Improved |
| add memory logit update (ablation) | X-DER · CIFAR-100 superclass | SS-NLL: 2.14 -> 1.83 | Improved |
| train from X-DER buffer logits instead of DER++ buffer | X-DER · Split CIFAR-100 and miniImageNet | buffer-only accuracy: 6% average gain (logits alone), 6.25% (with labels) | 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.
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 2 of 2 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
- Dark Experience for General Continual Learning: a Strong, Simple Baseline (NeurIPS 2020 · similarity 0.92)
- New Insights on Reducing Abrupt Representation Change in Online Continual Learning (ICLR 2022 · similarity 0.89)
- MIND: Multi-Task Incremental Network Distillation (AAAI 2024 · similarity 0.89)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.88)
- Rethinking Momentum Knowledge Distillation in Online Continual Learning (ICML 2024 · similarity 0.88)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.88)
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.
