Research map / Continual learning / Online and replay-based
Real-Time Evaluation in Online Continual Learning: A New Hope
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 |
|---|---|---|---|
| Use simple experience replay instead of considered continual-learning methods | CLOC fast-stream evaluation · CLOC | Average Online Accuracy: 11% | Helped |
| Match each expensive method's delay with additional ER gradient steps | ER, yielding ER-- · CLOC fast stream | Average Online Accuracy: up to 3.5% | Helped |
| Match compared methods' compute with additional ER gradient steps | ER, yielding ER++ · CLOC slow stream | Average Online Accuracy: 17%; 11%; 9.5% | Helped |
| Increase replay memory from 1×10^4 to 4×10^4 samples (ablation) | ER · CLOC fast stream | Average Online Accuracy: 18.09 -> 20.27 | Helped |
| Increase replay memory from 1×10^4 to 4×10^4 samples (ablation) | ACE · CLOC fast stream | Average Online Accuracy: 17.53 -> 19.42 | Helped |
| Increase replay memory from 1×10^4 to 4×10^4 samples (ablation) | LwF · CLOC fast stream | Average Online Accuracy: 16.87 -> 18.69 | Helped |
| Increase replay memory from 1×10^4 to 4×10^4 samples (ablation) | RWalk · CLOC fast stream | Average Online Accuracy: 14.60 -> 16.07 | Helped |
| Increase replay memory from 1×10^4 to 4×10^4 samples (ablation) | PoLRS · CLOC fast stream | Average Online Accuracy: 12.40 -> 13.59 | Helped |
| Increase replay memory from 1×10^4 to 4×10^4 samples (ablation) | MIR · CLOC fast stream | Average Online Accuracy: 11.37 -> 11.89 | Helped |
| Increase replay memory from 1×10^4 to 4×10^4 samples (ablation) | GSS · CLOC fast stream | Average Online Accuracy: 8.01 -> 9.10 | Helped |
| Use ACE loss rather than plain ER | Continual learner · CIFAR10 and CIFAR100 fast stream | Average Online Accuracy: 51.60 -> 53.05; 13.62 -> 14.42 | Mixed |
| Use two gradient-descent steps per time step for MIR and GSS (ablation) | MIR and GSS · CLOC slower stream | Average Online Accuracy: 17% to 14%; 11% to 10% | Helped |
Closest papers by meaning
- Budgeted Online Continual Learning by Adaptive Layer Freezing and Frequency-based Sampling (ICLR 2025 · similarity 0.90)
- CLDyB: Towards Dynamic Benchmarking for Continual Learning with Pre-trained Models (ICLR 2025 · similarity 0.90)
- Re-evaluating Continual Learning with Few-Shot Adaptation (NeurIPS 2026 · similarity 0.89)
- Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation (ICLR 2024 · similarity 0.88)
- Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning (NeurIPS 2024 · similarity 0.88)
- Where is the Truth? The Risk of Getting Confounded in a Continual World (ICML 2025 · 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.