Research map / Reinforcement learning / Visual and representation RL
Stochastic Neural Networks for Hierarchical Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 bilinear integration of observations and latent codes instead of concatenation (ablation) | SNN skill learning · Pre-training skill diversity | distinctive behaviors per latent code | Helped |
| Add a mutual-information reward bonus (ablation) | Bilinear SNN skill learning · Pre-training skill diversity | SNNs acquiring forward and backward motion: 80\% -> 100\% | Helped |
| Add a mutual-information reward bonus (ablation) | SNN skill learning · Pre-training skill diversity | skill diversity | Helped |
| Use bilinear SNN skills instead of a single Gaussian action policy | Downstream exploration · Locomotion + Maze and Food Gather | exploration coverage | Helped |
| Use a skill hierarchy instead of the CoM proxy reward alone | Downstream task learning · Locomotion + Maze | learning speed | Helped |
| Use the MI-pretrained SNN hierarchy instead of the hierarchy without MI pre-training (ablation) | SNN hierarchy · Locomotion + Maze | return | Mixed |
| Use the MI-pretrained SNN hierarchy instead of the hierarchy without MI pre-training (ablation) | SNN hierarchy · Food Gather | return | Helped |
| Use six independently trained policies instead of a single SNN | Hierarchical skill learning · Pre-training sample efficiency | samples required | Hurt |
| Use a 500-step switch interval instead of Gaussian action noise | Randomly initialized skill manager · Downstream exploration | CoM area explored | Helped |
| Vary the skill switch interval (ablation) | SNN hierarchy · Gather task | return | Mixed |
| Switch skills on an unstable robot | Ant skill hierarchy · Ant downstream tasks | rollout completion | Hurt |
| Use the fixed-horizon hierarchy on benchmark Gather settings with a 500-step path limit (ablation) | SNN hierarchy · Food Gather, benchmark settings | return | No clear effect |
Closest papers by meaning
- Probabilistic Subgoal Representations for Hierarchical Reinforcement Learning (ICML 2024 · similarity 0.88)
- Efficient Reinforcement Learning by Discovering Neural Pathways (NeurIPS 2024 · similarity 0.87)
- Improving Unsupervised Hierarchical Representation with Reinforcement Learning (CVPR 2024 · similarity 0.87)
- Deep Reinforcement Learning from Hierarchical Preference Design (ICML 2025 · similarity 0.87)
- Hierarchical Reinforcement Learning with Uncertainty-Guided Diffusional Subgoals (ICML 2025 · similarity 0.87)
- Stochastic Q-learning for Large Discrete Action Spaces (ICML 2024 · similarity 0.87)
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.