Research map / Continual learning / Online and replay-based
Online Continual Learning with Maximal Interfered Retrieval
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 |
|---|---|---|---|
| Replace random replay with MIR retrieval | Experience replay · MNIST Split | accuracy: 82.1\pm 1.5 -> 87.6 \pm 0.7 | Helped |
| Replace random replay with MIR retrieval | Experience replay · MNIST Split | forgetting: 15.0\pm2.1 -> 7.0 \pm 0.9 | Helped |
| Replace random replay with MIR retrieval | Experience replay · Permuted MNIST | accuracy: 78.9 \pm 0.6 -> 80.1 \pm 0.4 | Helped |
| Replace random replay with MIR retrieval | Experience replay · CIFAR-10 Split | accuracy: 27.5\pm 1.2 -> 29.8$\pm$1.1; 33.1\pm 1.7 -> 40.0 \pm 1.1; 41.3 \pm 1.9 -> 47.6 \pm 1.1 | Helped |
| Replace random replay with MIR retrieval | Experience replay · CIFAR-10 Split (M=100) | forgetting: 23.3 \pm 2.9 -> 17.4 \pm 2.1 | Helped |
| Use 5 iterations instead of 1 per incoming batch | Experience replay · CIFAR-10 Split (M=100) | accuracy: 41.3 \pm 1.9 -> 42.4 \pm 1.1 | Helped |
| Use 5 iterations instead of 1 per incoming batch | ER-MIR · CIFAR-10 Split (M=100) | accuracy: 47.6 \pm 1.1 -> 49.3 \pm 0.1 | Helped |
| Replace random replay with MIR retrieval | Experience replay · MiniImagenet Split | forgetting: 23.5\pm{1.0} -> 18.0\pm{0.8} | Helped |
| Replace random replay with MIR retrieval | Generative replay · MNIST Split | accuracy: 79.3 \pm 0.6 -> 82.1 \pm 0.3 | Helped |
| Replace random replay with MIR retrieval | Generative replay · Permuted MNIST | accuracy: 79.7 \pm 0.1 -> 80.4 \pm 0.2 | Helped |
| Replay maximally interfered samples rather than random samples | Generative model · MNIST Split | generator loss (negative ELBO): 107.2 \pm 0.2 -> 102.5 \pm 0.2 | Helped |
| Remove MIR retrieval from the generator (ablation) | GEN-MIR · MNIST Split | accuracy: 82.7 -> 83.0 | Helped |
Closest papers by meaning
- Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning (ICML 2024 · similarity 0.91)
- New Insights on Reducing Abrupt Representation Change in Online Continual Learning (ICLR 2022 · similarity 0.91)
- Summarizing Stream Data for Memory-Constrained Online Continual Learning (AAAI 2024 · similarity 0.90)
- Online Curvature-Aware Replay: Leveraging $\mathbf{2^{nd}}$ Order Information for Online Continual Learning (ICML 2025 · similarity 0.89)
- Predicting the Susceptibility of Examples to Catastrophic Forgetting (ICML 2025 · similarity 0.89)
- Budgeted Online Continual Learning by Adaptive Layer Freezing and Frequency-based Sampling (ICLR 2025 · similarity 0.89)
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.