Research map / Continual learning / Continual fine-tuning of LLMs
Achieving More with Less: Additive Prompt Tuning for Rehearsal-Free Class-Incremental Learning
Haoran Chen, Ping Wang, Zihan Zhou, Xu Zhang, Zuxuan Wu, Yu-Gang Jiang
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 progressive prompt fusion (PPF) at inference (ablation) | APT · Split CIFAR-100 (10 tasks) | Average accuracy: 88.17 -> 88.88 | Improved |
| add PPF at inference (ablation) | APT · Split ImageNet-R (10 tasks) | Average accuracy: 78.56 -> 79.40 | Improved |
| add PPF at inference (ablation) | APT · Split CIFAR-100 (10 tasks) | Forgetting: 7.05 -> 3.47 | Improved |
| add PPF at inference (ablation) | APT · Split ImageNet-R (10 tasks) | Forgetting: 8.83 -> 4.38 | Improved |
| add prompts to CLS key/value instead of CLS input (ablation) | APT · Split CIFAR-100 (10 tasks) | Average accuracy: 87.93 -> 88.88 | Improved |
| add prompts to CLS key/value instead of CLS input (ablation) | APT · Split ImageNet-R (10 tasks) | Average accuracy: 78.31 -> 79.40 | Improved |
| add prompts to CLS key/value instead of CLS input (ablation) | APT · Split CIFAR-100 (10 tasks) | Forgetting: 5.24 -> 3.47 | Improved |
| add prompts to CLS key/value instead of CLS input (ablation) | APT · Split ImageNet-R (10 tasks) | Forgetting: 5.18 -> 4.38 | No clear effect |
| vary PPF coefficient alpha from 0.2 to 0.8 (ablation) | APT · Split CIFAR-100 and ImageNet-R | Average accuracy | No clear effect |
| extend sequence from 10 to 20 tasks | APT · Split CIFAR-100 | Average accuracy: 88.88 -> 86.37 | Hurt |
| APT vs CPrompt on 20-task ImageNet-R forgetting | APT · Split ImageNet-R (20 tasks) | Forgetting: 6.41 -> 6.91 | Hurt |
| replace concatenated prompts with additive CLS prompts vs VPT-deep | APT · FGVC+VTAB average | Average accuracy: 84.4 -> 86.3 | Improved |
Closest papers by meaning
- CAPrompt: Cyclic Prompt Aggregation for Pre-Trained Model Based Class Incremental Learning (AAAI 2025 · similarity 0.91)
- PEARL: Input-Agnostic Prompt Enhancement with Negative Feedback Regulation for Class-Incremental Learning (AAAI 2025 · similarity 0.91)
- SEC-Prompt:SEmantic Complementary Prompting for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.91)
- OVOR: OnePrompt with Virtual Outlier Regularization for Rehearsal-Free Class-Incremental Learning (ICLR 2024 · similarity 0.88)
- CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning (CVPR 2025 · similarity 0.88)
- Harnessing Neural Unit Dynamics for Effective and Scalable Class-Incremental Learning (ICML 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.
