Research map / Continual learning / Online and replay-based
A Faster Path to Continual Learning
Wei Li, Hangjie Yuan, Zixiang Zhao, Borui Kang, Ziwei Liu, Tao Feng
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 |
|---|---|---|---|
| replace C-Flat with C-Flat Turbo | EASE · CIFAR100 B0_Inc10 | average accuracy: 91.91 -> 92.36 (C-Flat 92.05) | Improved |
| replace C-Flat with C-Flat Turbo | EASE · CIFAR100 B0_Inc10 | throughput (% of vanilla): 26.5% -> 61.6% | Improved |
| replace C-Flat with C-Flat Turbo | L2P · CIFAR100 B0_Inc10 | throughput (% of vanilla): 30.0% -> 59.4% | Improved |
| add C-Flat Turbo instead of C-Flat | MEMO · CIFAR100 from scratch ResNet-18 | last accuracy: 30.76 -> 32.24 | Improved |
| add C-Flat Turbo instead of C-Flat | iCaRL · CIFAR100 from scratch ResNet-18 | last accuracy: 42.47 -> 42.84 | Improved |
| add SAM to the base optimizer | L2P · CIFAR100 B0_Inc10 | average accuracy: 87.92 -> 88.18 | No clear effect |
| add LookSAM to the base optimizer | EASE · CIFAR100 B0_Inc10 | average accuracy: 91.16 -> 90.89 | Hurt |
| add the reuse of cached flatness direction (ablation) | EASE+C-Flat · CIFAR100 10-split | average accuracy: 92.15 -> 92.08; Img/s 44.25 -> 67.20 | Mixed |
| add linear step-size scheduler (ablation) | EASE+C-Flat Turbo reuse · CIFAR100 10-split | throughput (Img/s): 67.20 -> 74.63 | Improved |
| add adaptive trigger (ablation) | EASE+C-Flat Turbo · CIFAR100 10-split | average accuracy: 92.03 -> 92.25; Img/s 74.63 -> 102.74 | Improved |
| increase trigger threshold m from 1 to 2 (ablation) | C-Flat Turbo · CIFAR100 EASE | average accuracy: 92.06 -> 91.86; Img/s 102.74 -> 154.52 | Hurt |
| use trigger m=0.1 instead of 1 (ablation) | C-Flat Turbo · CIFAR100 EASE | throughput (Img/s): 102.74 -> 75.56 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 2 of 2 shared results
- Ranpac — RanPAC: Random Projections and Pre-trained Models for Continual Learning (NeurIPS 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- C-Flat++: Towards a More Efficient and Powerful Framework for Continual Learning (IJCV 2026 · similarity 0.92)
- Make Continual Learning Stronger via C-Flat (NeurIPS 2024 · similarity 0.89)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.87)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.87)
- An Effective Dynamic Gradient Calibration Method for Continual Learning (ICML 2024 · similarity 0.87)
- Convergence and Implicit Bias of Gradient Descent on Continual Linear Classification (ICLR 2025 · 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.
