Research map / Continual learning / Online and replay-based
Random Path Selection for Continual Learning
Jathushan Rajasegaran, Munawar Hayat, Salman H Khan, Fahad Shahbaz Khan, Ling Shao
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 |
|---|---|---|---|
| add distillation loss to single-path baseline (ablation) | RPS-Net baseline · CIFAR-100 | accuracy: 37.97% -> 44.93% | Improved |
| add controller phi(k, gamma) (ablation) | single path + distillation · CIFAR-100 | accuracy: 44.93% -> 51.76% | Improved |
| add multi-path random selection (ablation) | single path + distillation + controller · CIFAR-100 | accuracy: 51.76% -> 58.48% | Improved |
| use gamma=1 (iCaRL-style) instead of 2.5 (ablation) | RPS-Net · CIFAR-100 | accuracy | Hurt |
| use gamma=8 instead of 2.5 (ablation) | RPS-Net · CIFAR-100 | accuracy: 51% at task 10 | Mixed |
| increase number of paths (smaller J) (ablation) | RPS-Net · CIFAR-100 | accuracy | Hurt |
| vary number of modules M (including saturated M=1) (ablation) | RPS-Net · CIFAR-100 | accuracy | No clear effect |
| random path selection instead of genetic Binary Tournament Selection (ablation) | RPS-Net on MNIST · MNIST 5 tasks | average accuracy: 96.32% -> 96.52% | No clear effect |
| RPS-Net instead of iCARL | class-incremental CIFAR-100 · CIFAR-100 10 classes per task | accuracy: +7.3% | Improved |
| RPS-Net instead of iCARL | class-incremental ImageNet · ImageNet 100 classes, 10 per task | top-5 accuracy: +10.3% | Improved |
| RPS-Net instead of iCARL | backward transfer · CIFAR-100 | BWT: -0.4602 -> -0.1462 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- iCARL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 1 of 2 shared results
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
- Adaptive Progressive Continual Learning (TPAMI 2021 · similarity 0.87)
- Reinforced Continual Learning (NeurIPS 2018 · similarity 0.87)
- Fine-Grained Knowledge Selection and Restoration for Non-exemplar Class Incremental Learning (AAAI 2024 · similarity 0.85)
- Learning without Isolation: Pathway Protection for Continual Learning (ICML 2025 · similarity 0.84)
- Learning to Continually Learn with the Bayesian Principle (ICML 2024 · similarity 0.84)
- Overcoming Catastrophic Forgetting in Visual Continual Learning with Reinforcement Fine-Tuning (NeurIPS 2026 · similarity 0.84)
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.
