Research map / Continual learning / Online and replay-based
Is Parameter Isolation Better for Prompt-Based Continual Learning?
Jiangyang Li, Chenhao Ding, SongLin Dong, Qiang Wang, Jianchao Zhao, Yuhang He, Yihong Gong
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 history-aware dynamic routing (HDR) (ablation) | MoE prompt pool · Split CIFAR-100 (10 tasks) | FAA: 88.56 -> 94.15 | Improved |
| add history-aware dynamic routing (HDR) (ablation) | MoE prompt pool · Split ImageNet-R (10 tasks) | FAA: 76.99 -> 78.10 | Improved |
| add history-aware gradient modulation (HGM) (ablation) | MoE prompt pool · Split CIFAR-100 (10 tasks) | FAA: 88.56 -> 93.86 | Improved |
| add history-aware gradient modulation (HGM) (ablation) | MoE prompt pool · Split ImageNet-R (10 tasks) | FAA: 76.99 -> 77.56 | Improved |
| combine HDR and HGM (ablation) | MoE prompt pool · Split CIFAR-100 (10 tasks) | FAA: 88.56 -> 95.02 | Improved |
| use stepwise routing penalty instead of logarithmic (ablation) | Hash HDR · Split CIFAR-100 (10 tasks) | FAA: 91.02 -> 94.15 | Improved |
| use piecewise gradient modulation instead of inverse scaling (ablation) | Hash HGM · Split CIFAR-100 (10 tasks) | FAA: 92.45 -> 93.86 | Improved |
| deploy prompts at layers 1-4 instead of single shallow layer 1 (ablation) | Hash · Split CIFAR-100 (10 tasks) | FAA: 93.28 -> 95.02 | Improved |
| deploy prompts at layers 1-4 instead of layers 5-12 (ablation) | Hash · Split CIFAR-100 (10 tasks) | FAA: 94.67 -> 95.02 | Improved |
| increase router experts from 15 to 25 (ablation) | Hash · Split CIFAR-100 (10 tasks) | FAA: 95.02 -> 94.54 | Hurt |
| increase prompt length from 5 to 15 (ablation) | Hash · Split CIFAR-100 (10 tasks) | FAA: 94.12 -> 95.02 | Improved |
| share prompts across tasks with Hash instead of isolated prompts (NoRGa) | prompt-based CIL · Split ImageNet-R (10 tasks) | FAA: 75.40 -> 79.02 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DualPrompt — DualPrompt: Complementary Prompting for Rehearsal-Free Continual Learning (ECCV 2022) · 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
- Mixture of Experts Meets Prompt-Based Continual Learning (NeurIPS 2024 · similarity 0.93)
- Convolutional Prompting meets Language Models for Continual Learning (CVPR 2024 · similarity 0.92)
- Evolving Parameterized Prompt Memory for Continual Learning (AAAI 2024 · similarity 0.91)
- Training Consistent Mixture-of-Experts-Based Prompt Generator for Continual Learning (AAAI 2025 · similarity 0.91)
- InTAct: Interval-based Task Activation Consolidation for Continual Learning (NeurIPS 2026 · similarity 0.90)
- Consistent Prompting for Rehearsal-Free Continual Learning (CVPR 2024 · similarity 0.90)
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.
