Research map / Continual learning / Online and replay-based
TRGP: Trust Region Gradient Projection for Continual Learning
Sen Lin, Li Yang, Deliang Fan, Junshan Zhang
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 TRGP instead of GPM | continual learning · PMNIST | ACC (%): 93.91 -> 96.34 | Improved |
| use TRGP instead of GPM | continual learning · CIFAR-100 Split | ACC (%): 72.48 -> 74.46 | Improved |
| use TRGP instead of HAT | continual learning · 5-Dataset | ACC (%): 91.32 -> 93.56 | Improved |
| use TRGP instead of GPM | continual learning · MiniImageNet | ACC (%): 60.41 -> 61.78 | Improved |
| use TRGP instead of GPM | continual learning · CIFAR-100 Sup | ACC (%): 57.72 -> 58.25 | Improved |
| vary εˡ among 0.2, 0.5, and 0.7 (ablation) | TRGP · CIFAR-100 Split | ACC: 74.52, 74.46, 74.30 | No clear effect |
| vary εˡ among 0.2, 0.5, and 0.7 (ablation) | TRGP · 5-Dataset | ACC: 93.28, 93.56, 93.43 | No clear effect |
| use a fixed task-wise trust region across layers instead of layer-wise selection (ablation) | TRGP · CIFAR-100 Split | ACC: 74.46 -> 73.25 | Hurt |
| use a fixed task-wise trust region across layers instead of layer-wise selection (ablation) | TRGP · 5-Dataset | ACC: 93.56 -> 92.85 | Hurt |
| select top-2 correlated old tasks instead of top-1 (ablation) | TRGP · CIFAR-100 Split | ACC: 74.00 -> 74.46 | Improved |
| select top-2 correlated old tasks instead of top-1 (ablation) | TRGP · 5-Dataset | ACC: 92.94 -> 93.56 | Improved |
Later papers that compared with this method
10 later accepted papers report TRGP as a baseline in their results tables; 9 of them report a better number on most of the results they share with it.
- Plasticity Activation via Polar Operator: A Plug-in Method for Balancing Stability and Plasticity (ICML 2026 · better on 4 of 4 shared results)
- LANCE: Low Rank Activation Compression for Efficient On-Device Continual Learning (ECCV 2026 · better on 0 of 1 shared result)
- BECAME: Bayesian Continual Learning with Adaptive Model Merging (ICML 2025 · better on 3 of 3 shared results)
- CODE-CL: Conceptor-Based Gradient Projection for Deep Continual Learning (ICCV 2025 · better on 3 of 3 shared results)
- Adaptive Plasticity Improvement for Continual Learning (CVPR 2023 · better on 4 of 5 shared results)
- Rethinking Gradient Projection Continual Learning: Stability / Plasticity Feature Space Decoupling (CVPR 2023 · better on 4 of 4 shared results)
- Theory on Forgetting and Generalization of Continual Learning (ICML 2023 · better on 4 of 4 shared results)
- Continual Learning with Scaled Gradient Projection (AAAI 2023 · better on 2 of 2 shared results)
2 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
- Trust Region Continual Learning as an Implicit Meta-Learner (NeurIPS 2026 · similarity 0.88)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.87)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.87)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.87)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.87)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.86)
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.
