Research map / Continual learning / Online and replay-based
Beyond Not-Forgetting: Continual Learning with Backward Knowledge Transfer
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 |
|---|---|---|---|
| Add selective backward-transfer model updates (CUBER) over TRGP | CUBER · Split CIFAR-100 | BWT (%): -0.9 -> 0.13 | Helped |
| CUBER vs TRGP | CUBER · Split MiniImageNet | ACC (%): 61.78 -> 62.67 | Helped |
| CUBER vs TRGP | CUBER · Permuted MNIST | ACC (%): 96.34 -> 97.25 | Helped |
| CUBER vs TRGP on less-correlated tasks | CUBER · 5-Dataset | ACC (%): 93.56 -> 93.48 | No clear effect |
| CUBER vs TRGP on OL-CIFAR100 | CUBER · OL-CIFAR100 | ACC (%): 74.33 -> 74.94 | Helped |
| CUBER vs TRGP on OL-CIFAR100 | CUBER · OL-CIFAR100 | BWT (%): -0.18 -> 0.28 | Helped |
| Task selection by Regime 3 (CUBER vs TRGP) | CUBER · OL-CIFAR100 | BWT-S average: -0.23 -> 0.33 | Helped |
| Raise epsilon_2 from 0.0 to 0.2 (ablation) | CUBER · Split CIFAR-100 | BWT (%): 0.22 -> 0.03 | Hurt |
| Raise epsilon_2 from 0.0 to 0.5 (ablation) | CUBER · Split CIFAR-100 | BWT (%): 0.22 -> 0.01 | Hurt |
| Vary epsilon_2 (0.0, 0.2, 0.5) (ablation) | CUBER · Split CIFAR-100 | ACC (%): 75.54 -> 75.73 | No clear effect |
| Relax gradient constraint via Regime 3 (CUBER vs TRGP) | CUBER · Split CIFAR-100 | FWT gain over GPM: 2.01 -> 2.79 | Helped |
| CUBER vs TRGP forward transfer on 5-Dataset | CUBER · 5-Dataset | FWT gain over GPM: 1.98 -> 1.96 | No clear effect |
Closest papers by meaning
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.88)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.88)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.88)
- Do Your Best and Get Enough Rest for Continual Learning (CVPR 2025 · similarity 0.88)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.88)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 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.