Research map / Continual learning / Continual fine-tuning of LLMs
DualPrompt: Complementary Prompting for Rehearsal-Free Continual Learning
Zifeng Wang, Zizhao Zhang, Sayna Ebrahimi, Ruoxi Sun, Han Zhang, Chen-Yu Lee, Xiaoqi Ren, Guolong Su, Vincent Perot, Jennifer Dy, Tomas Pfister
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 |
|---|---|---|---|
| use Prefix Tuning instead of Prompt Tuning (ablation) | DualPrompt · Split CIFAR-100 | average accuracy: 83.81 -> 86.51 | Improved |
| use Prefix Tuning instead of Prompt Tuning (ablation) | DualPrompt · Split ImageNet-R | average accuracy: 64.99 -> 68.13 | Improved |
| add G-Prompt alone to frozen backbone (ablation) | DualPrompt · Split ImageNet-R | average accuracy: 27.01 -> 63.41 | Improved |
| add E-Prompt alone to frozen backbone (ablation) | DualPrompt · Split ImageNet-R | average accuracy: 27.01 -> 65.10 | Improved |
| add G-Prompt to E-Prompt (ablation) | DualPrompt · Split ImageNet-R | average accuracy: 65.10 -> 66.77 | Improved |
| add E-Prompt to G-Prompt (ablation) | DualPrompt · Split ImageNet-R | forgetting: 6.52 -> 5.74 | Improved |
| extend G- and E-Prompt to multiple layers (ablation) | DualPrompt · Split ImageNet-R | average accuracy: 66.77 -> 68.13 | Improved |
| extend G-Prompt alone to multiple layers (ablation) | DualPrompt G-Prompt only · Split ImageNet-R | forgetting: 63.41 -> 63.85 | Mixed |
| attach G-Prompt to layers 1-12 instead of 1-2 (ablation) | DualPrompt · Split ImageNet-R (validation) | average accuracy: 68.46 -> 63.13 | Hurt |
| use ground-truth task identity to select E-Prompt (ablation) | DualPrompt · Split ImageNet-R | average accuracy: 68.13 -> 71.97 | Improved |
| replace L2P with DualPrompt | DualPrompt · Split ImageNet-R | forgetting: 9.73 -> 4.68 | Improved |
| replace L2P with DualPrompt | DualPrompt · Split CIFAR-100 | average accuracy: 83.86 -> 86.51 | Improved |
Later papers that compared with this method
50 later accepted papers report DualPrompt as a baseline in their results tables; 49 of them report a better number on most of the results they share with it.
- InTAct: Interval-based Task Activation Consolidation for Continual Learning (NeurIPS 2026 · better on 5 of 5 shared results)
- Quantized Residuals to Continuous Prompts for Few-Shot Class Incremental Learning in Vision-Language Models (CVPR 2026 · better on 3 of 4 shared results)
- DeLo: Dual Decomposed Low-Rank Experts Collaboration for Continual Missing Modality Learning (AAAI 2026 · better on 4 of 4 shared results)
- Learning Probabilistic Prompt for Continual Learning (ECCV 2026 · better on 3 of 3 shared results)
- One-Prompt Strikes Back: Sparse Mixture of Experts for Prompt-based Continual Learning (ICLR 2026 · better on 2 of 2 shared results)
- DK-DDIL: Adaptive Knowledge Retention for Dynamic Domain-Incremental Learning in Medical Imaging (CVPR 2026 · better on 2 of 2 shared results)
- Learning Adaptive and Expandable Mixture Model for Continual Learning (AAAI 2026 · better on 2 of 2 shared results)
- Beyond Point-wise Neural Collapse: A Topology-Aware Hierarchical Classifier for Class-Incremental Learning (ICML 2026 · better on 1 of 1 shared result)
42 more not listed.
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
- Advancing Prompt-Based Methods for Replay-Independent General Continual Learning (ICLR 2025 · similarity 0.89)
- Unlocking the Power of Rehearsal in Continual Learning: A Theoretical Perspective (ICML 2025 · similarity 0.89)
- Mixture of Experts Meets Prompt-Based Continual Learning (NeurIPS 2024 · similarity 0.89)
- Consistent Prompting for Rehearsal-Free Continual Learning (CVPR 2024 · similarity 0.89)
- REP: Resource-Efficient Prompting for Rehearsal-Free Continual Learning (NeurIPS 2025 · similarity 0.88)
- Convolutional Prompting meets Language Models for Continual Learning (CVPR 2024 · 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.
