Research map / Continual learning / Online and replay-based
Neuro-evolutionary Continual Reinforcement Learning
Pengyi Li, Hongyao Tang, Yifu Yuan, YAN ZHENG, Xin Xu, Jianye HAO
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 |
|---|---|---|---|
| replace semantic-similarity mask construction with random mask construction (ablation) | Nevo-CRL · CW10 | average performance | Hurt |
| add population of masks and policies versus single mask and single RL individual (ablation) | Nevo-CRL · CW10 | average performance | Improved |
| replace random crossover with importance-guided crossover (ablation) | Nevo-CRL · CW10 | average performance | Improved |
| add pruning of low-activity connections (ablation) | Nevo-CRL · CW10 | average performance | No clear effect |
| add pruning of low-activity connections (ablation) | Nevo-CRL · CW10 | capacity utilization: 100% after a few tasks without pruning | Improved |
| use Nevo-CRL instead of CoTASP | CoTASP · CW10 | average performance P: 0.89 -> 0.96 | Improved |
| use Nevo-CRL instead of CoTASP | CoTASP · CW20 | average performance P: 0.86 -> 0.93 | Improved |
| use Nevo-CRL instead of PackNet | PackNet · CW10 | generalization G: 0.38 -> 0.41 | Hurt |
| use Nevo-CRL instead of CoTASP | CoTASP · CW10 | generalization G: 0.47 -> 0.41 | Improved |
| use Nevo-CRL instead of CoTASP on Brax HalfCheetah transfer | CoTASP · Brax HalfCheetah transfer | average return: 780.71 -> 3211.93 | Improved |
| use Nevo-CRL instead of CoTASP on Brax Ant forgetting | CoTASP · Brax Ant forgetting | average return: 2746 -> 4003 | Improved |
| use population-based training instead of single policy (computational cost) | CoTASP · CW20 | training time: 45.9 -> 70.8 hours | Hurt |
Closest papers by meaning
- Rewiring Neurons in Non-Stationary Environments (NeurIPS 2023 · similarity 0.89)
- Stay Hungry, Keep Learning: Sustainable Plasticity for Deep Reinforcement Learning (ICML 2025 · similarity 0.88)
- Self-Composing Policies for Scalable Continual Reinforcement Learning (ICML 2024 · similarity 0.87)
- Continual Knowledge Adaptation for Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.87)
- InTAct: Interval-based Task Activation Consolidation for Continual Learning (NeurIPS 2026 · 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.
