Research map / Reinforcement learning / Visual and representation RL
Creating Multi-Level Skill Hierarchies in Reinforcement Learning
Joshua Benjamin Evans, Özgür Şimşek
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 the Louvain skill hierarchy instead of alternative skill-discovery methods or primitive actions | Reinforcement-learning agents · Rooms, Grid, Maze, Office, Taxi | learning performance | Improved |
| Use the Louvain skill hierarchy instead of other hierarchical agents or primitive actions | Reinforcement-learning agents · Towers of Hanoi | learning performance | No clear effect |
| Use Louvain skills in a multi-level hierarchy rather than a flat arrangement that calls primitive actions directly (ablation) | Louvain skills · Rooms, Taxi, Maze | learning performance | Improved |
| Use the full Louvain hierarchy rather than skills from only one hierarchy level (ablation) | Louvain skills · Rooms, Taxi, Maze | learning performance | Improved |
| Increase the Office environment size and use Louvain skills | Louvain agent · Multi-floor Office, 2537 states | learning performance | Improved |
| Use Louvain skills generated with lower rather than higher resolution parameter values (ablation) | Louvain agent · Rooms, Towers of Hanoi | learning performance | Improved |
| Use very high resolution values (ρ ≥ 10) rather than the full Louvain hierarchy (ablation) | Louvain agent · Rooms, Towers of Hanoi | learning performance | Hurt |
| Incrementally discover Louvain skills rather than use primitive actions only | Incremental Louvain agent · Rooms | learning performance | Improved |
| Replace the skill hierarchy from scratch at graph updates rather than update the existing hierarchy (ablation) | Incremental Louvain agent · Rooms | asymptotic learning performance | Improved |
| Update the existing hierarchy while keeping existing nodes' cluster membership fixed rather than rebuilding from scratch (ablation) | Incremental Louvain agent · Rooms | learning performance | No clear effect |
Closest papers by meaning
- Hierarchical Multi-Agent Skill Discovery (NeurIPS 2023 · similarity 0.90)
- Hierarchical and Interpretable Skill Acquisition in Multi-task Reinforcement Learning (ICLR 2018 · similarity 0.90)
- Offline Multi-agent Continual Cooperation via Skill Partition and Reuse (ICML 2026 · similarity 0.89)
- When Do Skills Help Reinforcement Learning? A Theoretical Analysis of Temporal Abstractions (ICML 2024 · similarity 0.88)
- Learning Generalizable Skills from Offline Multi-Task Data for Multi-Agent Cooperation (ICLR 2025 · similarity 0.87)
- Skill Machines: Temporal Logic Skill Composition in Reinforcement Learning (ICLR 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.
