Research map / Continual learning / Online and replay-based
Navigating Memory Construction by Global Pseudo-Task Simulation for Continual Learning
Yejia Liu, Wang Zhu, Shaolei Ren
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 permutation pseudo-tasks to simulate memory construction (vs ER-Hybrid) | GPS · P-MNIST | accuracy: 86.84 -> 87.93 | Improved |
| Use permutation pseudo-tasks (vs ER-Hybrid) | GPS · TinyImageNet | accuracy: 57.97 -> 59.26 | Improved |
| Replace permutation with rotation simulation (ablation) | GPS · P-MNIST | accuracy: 87.93 -> 85.38 | Hurt |
| Replace permutation with blurring simulation (ablation) | GPS · TinyImageNet | accuracy: 59.26 -> 56.85 | Hurt |
| Replace permutation with rotation simulation (ablation) | GPS · S-CIFAR-10 | accuracy: 92.77 -> 91.61 | Hurt |
| Offline oracle versus GPS permutation | ER-Oracle · TinyImageNet | accuracy: 59.26 -> 60.56 | Improved |
| Add GPS memory construction to DER++ | DER++ · P-MNIST | accuracy: 91.14 -> 91.64 | Improved |
| Add GPS to HAL | HAL · P-MNIST | accuracy: 87.69 -> 88.23 | Improved |
| Add GPS to DER++ | DER++ · TinyImageNet | accuracy: 60.67 -> 61.01 | Improved |
| Use GPS with small memory buffer (vs ER-Hybrid) | GPS · S-CIFAR-10, memory 20 | accuracy | No clear effect |
| Use GPS with curriculum policies (vs ER-CurRes) | GPS w/ Cur · S-CIFAR-100, memory 2000 | accuracy: 81.38 -> 82.34 | Improved |
| Use GPS with curriculum policies (vs ER-CurRes) | GPS w/ Cur · TinyImageNet, memory 2000 | accuracy: 58.89 -> 59.88 | Improved |
Closest papers by meaning
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.91)
- Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning (ICML 2024 · similarity 0.90)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.90)
- Predicting the Susceptibility of Examples to Catastrophic Forgetting (ICML 2025 · similarity 0.90)
- Forget Forgetting: Continual Learning in a World of Abundant Memory (ICLR 2026 · similarity 0.90)
- Advancing Prompt-Based Methods for Replay-Independent General Continual Learning (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.
