Research map / Continual learning / Continual fine-tuning of LLMs
Plasticity Activation via Polar Operator: A Plug-in Method for Balancing Stability and Plasticity
Guodong Zheng, Enneng Yang, Xiaoyan Wang, Yihan Chen, Feihong He, Quan Zheng, Peng Wang, Li Shen
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 PAPO (polar operator gradient rectification) to ER | ER · MiniImageNet (20 tasks) | ACC (%): 63.40 -> 72.41 | Improved |
| add PAPO to GPM | GPM · MiniImageNet (20 tasks) | ACC (%): 60.20 -> 69.10 | Improved |
| add PAPO to TRGP | TRGP · MiniImageNet (20 tasks) | ACC (%): 61.78 -> 70.97 | Improved |
| add PAPO to ER on P-MNIST | ER · P-MNIST (10 tasks) | ACC (%): 87.24 -> 88.18 | Improved |
| add PAPO to TRGP on P-MNIST | TRGP · P-MNIST (10 tasks) | ACC (%): 96.34 -> 96.73 | Improved |
| add PAPO to InfLoRA | InfLoRA · SuperNI Order 1 | ACC: 42.93 -> 48.08 | Improved |
| add PAPO to InfLoRA | InfLoRA · SuperNI Order 1 | forgetting (FT): 11.23 -> 6.29 | Improved |
| add PAPO to GainLoRA | GainLoRA · SuperNI Order 2 | ACC: 50.17 -> 51.45 | Improved |
| apply PAPO every 30 epochs instead of every 1-5 epochs (ablation) | ER+PAPO · MiniImageNet | ACC | Hurt |
| apply PAPO at every epoch (s=1) (ablation) | PAPO · MiniImageNet | ACC and BWT | No clear effect |
| use 1 to 4 Polar Express iterations instead of zero (ablation) | PAPO · MiniImageNet | ACC and BWT | Improved |
| increase balance coefficient lambda to 1 or more (ablation) | GPM+PAPO, TRGP+PAPO · MiniImageNet | ACC and BWT | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TRGP — TRGP: Trust Region Gradient Projection for Continual Learning (ICLR 2022) · better on 4 of 4 shared results
- InfLoRA — InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning (CVPR 2024) · better on 3 of 3 shared results
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
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.89)
- Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective (ICML 2025 · similarity 0.89)
- Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning (ICLR 2024 · similarity 0.89)
- Learning Continually by Spectral Regularization (ICLR 2025 · similarity 0.88)
- Self-Normalized Resets for Plasticity in Continual Learning (ICLR 2025 · similarity 0.88)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · 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.
