Research map / Continual learning / Online and replay-based
Dark Experience for General Continual Learning: a Strong, Simple Baseline
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 |
|---|---|---|---|
| Replay stored logits with the DER++ objective rather than replaying examples alone | ER-style continual learning · Sequential CIFAR-10 Class-IL, buffer 200 | accuracy: 44.79 -> 64.88 | Helped |
| Replay stored logits with the DER++ objective rather than replaying examples alone | ER-style continual learning · Sequential Tiny ImageNet Class-IL, buffer 200 | accuracy: 8.49 -> 10.96 | Helped |
| Replay stored logits with the DER++ objective rather than replaying examples alone | ER-style continual learning · Permuted MNIST Domain-IL, buffer 200 | accuracy: 72.37 -> 83.58 | Helped |
| Replay stored logits with the DER++ objective rather than replaying examples alone | ER-style continual learning · Rotated MNIST Domain-IL, buffer 200 | accuracy: 85.01 -> 90.43 | Helped |
| Add a buffer-label classification loss to DER (ablation) | DER · Sequential CIFAR-10 Class-IL, buffer 200 | accuracy: 61.93 -> 64.88 | Helped |
| Add a buffer-label classification loss to DER (ablation) | DER · Sequential Tiny ImageNet Class-IL, buffer 200 | accuracy: 11.87 -> 10.96 | Hurt |
| Add a buffer-label classification loss to DER (ablation) | DER · MNIST-360 test, buffer 200 | accuracy: 55.22 -> 54.16 | Hurt |
| Sample stored logits throughout the optimization trajectory instead of only at task boundaries (ablation) | DER and DER++ · Sequential MNIST Class-IL | weighted cumulative Fisher energy | Helped |
Closest papers by meaning
- Advancing Prompt-Based Methods for Replay-Independent General Continual Learning (ICLR 2025 · similarity 0.90)
- Continual evaluation for lifelong learning: Identifying the stability gap (ICLR 2023 · similarity 0.89)
- GACL: Exemplar-Free Generalized Analytic Continual Learning (NeurIPS 2024 · similarity 0.87)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.87)
- Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning (ICML 2024 · similarity 0.87)
- A Unified and General Framework for Continual Learning (ICLR 2024 · 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.