Research map / Continual learning / Continual fine-tuning of LLMs
Parameter-Efficient Fine-Tuning for Continual Learning: A Neural Tangent Kernel Perspective
Jingren Liu, Zhong Ji, Yunlong Yu, Jiale Cao, Yanwei Pang, Jungong Han, Xuelong Li
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 knowledge retention (KR) to S2 (ablation) | S2 adapter · CIFAR-100 (IN21K) | incremental accuracy (%): 85.14 -> 91.29 | Improved |
| add knowledge retention (KR) to S1 (ablation) | S1 adapter · CIFAR-100 (IN21K) | incremental accuracy (%): 83.22 -> 84.61 | Improved |
| add knowledge retention (KR) to S1+S2+Hybrid (ablation) | NTK-CL · CIFAR-100 (IN21K) | incremental accuracy (%): 89.18 -> 91.89 | Improved |
| add knowledge retention (KR) to S1+S2+Hybrid (ablation) | NTK-CL · ImageNet-R (IN21K) | incremental accuracy (%): 75.94 -> 80.91 | Improved |
| use hybrid adaptation instead of S1 only (ablation) | single-branch adapter · CIFAR-100 (IN21K) | incremental accuracy (%): 83.22 -> 86.75 | Improved |
| add task-feature dissimilarity loss (ablation) | NTK-CL with KR · CIFAR-100 (IN21K) | incremental accuracy (%): 91.89 -> 92.83 | Improved |
| add orthogonality loss (ablation) | NTK-CL with KR · CIFAR-100 (IN21K) | incremental accuracy (%): 91.89 -> 92.39 | Improved |
| add regularization loss (ablation) | NTK-CL with KR · CIFAR-100 (IN21K) | incremental accuracy (%): 91.89 -> 92.16 | Improved |
| add regularization loss to full constraints (ablation) | NTK-CL · CIFAR-100 (IN21K) | incremental accuracy (%): 93.06 -> 93.76 | Improved |
| add regularization loss to full constraints (ablation) | NTK-CL · ImageNet-R (IN21K) | incremental accuracy (%): 82.50 -> 82.77 | Improved |
| NTK-CL versus best baseline on ImageNet-A | NTK-CL · ImageNet-A (IN21K) | incremental accuracy (%): 64.35 -> 66.56 | Improved |
| NTK-CL versus OVOR on DomainNet | NTK-CL · DomainNet (IN21K) | incremental accuracy (%): 75.80 -> 73.70 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- OVOR — OVOR: OnePrompt with Virtual Outlier Regularization for Rehearsal-Free Class-Incremental Learning (ICLR 2024) · better on 2 of 3 shared results
- InfLoRA — InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024) · better on 1 of 1 shared result
- 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
- SAFE: Slow and Fast Parameter-Efficient Tuning for Continual Learning with Pre-Trained Models (NeurIPS 2024 · similarity 0.91)
- BiLoRA: Almost-Orthogonal Parameter Spaces for Continual Learning (CVPR 2025 · similarity 0.90)
- Overcoming Catastrophic Forgetting in Visual Continual Learning with Reinforcement Fine-Tuning (NeurIPS 2026 · similarity 0.90)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.90)
- The Importance of Being Lazy: Scaling Limits of Continual Learning (ICML 2025 · similarity 0.89)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.89)
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.
