Research map / Continual learning / Online and replay-based
Parameter-efficient Continual Learning for Enhancing Plasticity without Forgetting under Limited Model Capacity
Yitian Chen, Shigeng Zhang, Xuan Liu, Mingming Lu, Kai Chen, Hongye Zhu, Xinning Chen
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 gradient-guided parameter reuse (vs reusing all frozen parameters) (ablation) | GRAPA · MFKT R2 (ResNet-18) | mean test accuracy (%): 62.79 -> 85.26 | Improved |
| add gradient-guided parameter reuse (vs reusing all frozen parameters) (ablation) | GRAPA · QSCF R1 (ResNet-18) | mean test accuracy (%): 81.29 -> 90.25 | Improved |
| add gradient-guided parameter reuse (vs reusing all frozen parameters) (ablation) | GRAPA · MFKT C (CNN) | mean test accuracy (%): 74.55 -> 76.54 | Improved |
| add adaptive RL-based allocation (vs fixed pruning rate) (ablation) | GRAPA · MFKT R1 (ResNet-18) | mean test accuracy (%): 73.04 -> 85.89 | Improved |
| add adaptive RL-based allocation (vs fixed pruning rate) (ablation) | GRAPA · QSCF R1 (ResNet-18) | mean test accuracy (%): 71.00 -> 90.25 | Improved |
| add adaptive RL-based allocation (vs fixed pruning rate) (ablation) | GRAPA · MFKT C (CNN) | mean test accuracy (%): 73.73 -> 76.54 | Improved |
| use RL-based pruning rate instead of linear-decay pruning (ablation) | GRAPA · QSCF R1 (ResNet-18) | mean test accuracy (%): 79.58 -> 90.25 | Improved |
| use RL-based pruning rate instead of binary-search pruning (ablation) | GRAPA · QSCF R1 (ResNet-18) | mean test accuracy (%): 83.04 -> 90.25 | Improved |
| increase alpha with small beta in reward (ablation) | GRAPA · MFKT R1 (ResNet-18) | accuracy | Hurt |
| vary reward alpha and beta (ablation) | GRAPA · MFKT C (CNN) | accuracy | No clear effect |
| GRAPA vs PackNet (fixed pruning rate isolation) | GRAPA · QSCF R1 (ResNet-18) | mean test accuracy (%): 80.64 -> 90.25 | Improved |
| GRAPA vs Independent full models | GRAPA · MFKT C (CNN) | mean test accuracy (%): 76.86 vs 76.54 | No clear effect |
Closest papers by meaning
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.89)
- Pareto Continual Learning: Preference-Conditioned Learning and Adaption for Dynamic Stability-Plasticity Trade-off (AAAI 2025 · similarity 0.89)
- BECAME: Bayesian Continual Learning with Adaptive Model Merging (ICML 2025 · similarity 0.88)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.88)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.88)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 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.
