Research map / Continual learning / Online and replay-based
SPHERE: Mitigating the Loss of Spectral Plasticity in Mixture-of-Experts for Deep Reinforcement Learning
Lirui Luo, Guoxi Zhang, Hongming Xu, Cong Fang, Qing 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 the SPHERE Parseval penalty on last-layer weighted expert features (ablation) | Top-K MoE PPO · HumanoidBench CRL | average success: 0.36 -> 0.54 | Improved |
| add SPHERE to Top-K MoE | Top-K MoE PPO · MetaWorld CRL | average success: +133% | Improved |
| add SPHERE to Top-K MoE | Top-K MoE PPO · HumanoidBench RL | average success: +36% | Improved |
| add SPHERE to Top-K MoE | Top-K MoE PPO · HumanoidBench CRL | average success: +50% | Improved |
| regularize only the last hidden expert layer instead of all hidden expert layers (ablation) | SPHERE · HumanoidBench CRL | average success: 0.42 -> 0.54 | Improved |
| add cross-expert concatenation instead of per-expert loss sum (ablation) | SPHERE · HumanoidBench CRL | average success: 0.40 -> 0.54 | Improved |
| regularize feature Gram A_last instead of gradient Gram G_last (ablation) | SPHERE · HumanoidBench CRL | average success: 0.43 -> 0.54 | Improved |
| switch from RL to CRL | PPO and MoE policies · HumanoidBench | average success | Hurt |
Closest papers by meaning
- A Study of Plasticity Loss in On-Policy Deep Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Mixtures of Experts Unlock Parameter Scaling for Deep RL (ICML 2024 · similarity 0.89)
- Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churn (ICML 2025 · similarity 0.87)
- Revisiting Plasticity in Visual Reinforcement Learning: Data, Modules and Training Stages (ICLR 2024 · similarity 0.87)
- Learning Continually by Spectral Regularization (ICLR 2025 · similarity 0.86)
- Mixture-of-Experts Meets In-Context Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
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.
