Research map / Reinforcement learning / Visual and representation RL
Self-Supervised Mixture-of-Experts by Uncertainty Estimation
Zhuobin Zheng, Chun Yuan, Xinrui Zhu, Zhihui Lin, Yangyang Cheng, Cheng Shi, Jiahui Ye
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 SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah SmallFoot (fully trained) | cumulative reward: 2003.46 -> 3460 | Improved |
| use SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah BigFoot (fully trained) | cumulative reward: 2211.92 -> 2392 | Improved |
| use SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah SmallLeg (fully trained) | cumulative reward: 2327.16 -> 4233 | Improved |
| use SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah BigLeg (fully trained) | cumulative reward: 2269.78 -> 3147 | Improved |
| use SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah SmallThigh (fully trained) | cumulative reward: 2555.16 -> 3346 | Improved |
| use SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah BigThigh (fully trained) | cumulative reward: 2424.95 -> 2465 | Improved |
| use SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah SmallTorso (fully trained) | cumulative reward: 2294.72 -> 2870 | Improved |
| use SUM (K=3) instead of TRPO | multi-task agent · HalfCheetah BigTorso (fully trained) | cumulative reward: 2686.13 -> 3478 | Improved |
| train with NLL uncertainty objective instead of MSE, single head K=1 (ablation) | DDPG · Walker and Hopper | averaged return | Improved |
| add uncertainty enhancement with K=3 heads versus multi-head DDPG (ablation) | SUM (K=3) · Walker/Hopper | averaged return and learning speed | Improved |
| uncertainty-trained critic versus multi-head DDPG, generalization gap (ablation) | SUM · Hopper and Walker test vs train | performance decrease factor alpha: [0.32, 0.51] -> [0.05, 0.28] | Improved |
| self-supervised expert scheduling versus SOUP confidence strategy | SUM · Hopper and Humanoid multi-task | averaged return | Improved |
Closest papers by meaning
- Mastering Massive Multi-Task Reinforcement Learning via Mixture-of-Expert Decision Transformer (ICML 2025 · similarity 0.89)
- Multi-Task Reinforcement Learning with Mixture of Orthogonal Experts (ICLR 2024 · similarity 0.89)
- Mixtures of Experts Unlock Parameter Scaling for Deep RL (ICML 2024 · similarity 0.88)
- Mixture-of-Experts Meets In-Context Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Learning to Specialize: Joint Gating-Expert Training for Adaptive MoEs in Decentralized Settings (NeurIPS 2025 · similarity 0.86)
- Autonomy-of-Experts Models (ICML 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.
