Research map / Continual learning / Online and replay-based
PROL : Rehearsal Free Continual Learning in Streaming Data via Prompt Online Learning
M. Anwar Ma'sum, Mahardhika Pratama, Savitha Ramasamy, Lin Liu, Habibullah Habibullah, Ryszard Kowalczyk
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 prompt generator G to fine-tuned backbone (ablation) | FT · ImageNet-R | FAA: 12.68 -> 69.77 | Improved |
| add inter-task and intra-task loss together vs inter only (ablation) | FT · ImageNet-R | FAA: 11.98 -> 12.68 | Improved |
| add scaler-shifters, class keys and similarity loss (ablation) | FT+G · ImageNet-R | FAA: 69.77 -> 70.82 | Improved |
| add orthogonality loss on keys (ablation) | FT+G+SS+K · ImageNet-R | FAA: 70.82 -> 71.13 | Improved |
| add hard-soft update strategy (ablation) | FT+G+SS+K · ImageNet-R | FAA: 71.13 -> 72.53 | Improved |
| add hard-soft update strategy (ablation) | FT+G+SS+K · ImageNet-R | FM: 4.31 -> 4.95 | Hurt |
| add PTM generalization preservation (M and L_gen) (ablation) | PROL w/o M · ImageNet-R | FAA: 72.53 -> 73.50 | Improved |
| PROL vs ConvPrompt on CIFAR100 | ConvPrompt · CIFAR100 | FAA: 84.79 -> 86.32 | Improved |
| PROL vs ViT-Tiny vs non-PTM ERSM | ERSM · CIFAR100 | FAA: 47.72 -> 64.59 | Improved |
| PROL vs RanPAC-J on CUB200 | RanPAC-J · CUB200 | FAA: 85.20 -> 72.51 | Hurt |
| vary prompt length from 3 to 15 (ablation) | PROL · ImageNet-R | FAA: 73.42 -> 73.30 | No clear effect |
| PROL throughput vs ConvPrompt | ConvPrompt · ImageNet-R | throughput: 11.77 -> 35.42 | Improved |
Closest papers by meaning
- Advancing Prompt-Based Methods for Replay-Independent General Continual Learning (ICLR 2025 · similarity 0.90)
- One-stage Prompt-based Continual Learning (ECCV 2024 · similarity 0.90)
- REP: Resource-Efficient Prompting for Rehearsal-Free Continual Learning (NeurIPS 2025 · similarity 0.89)
- Summarizing Stream Data for Memory-Constrained Online Continual Learning (AAAI 2024 · similarity 0.89)
- Federated Continual Learning Goes Online: Uncertainty-Aware Memory Management for Vision Tasks and Beyond (ICLR 2025 · similarity 0.87)
- Training Consistent Mixture-of-Experts-Based Prompt Generator for Continual Learning (AAAI 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.
