Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Options via Compression
Yiding Jiang, Evan Zheran Liu, Benjamin Eysenbach, J Zico Kolter, Chelsea Finn
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 trajectory compression to learn options instead of maximizing likelihood alone | Offline skill learning · Grid world demonstrations | F1: 0.32 -> 0.92 | Improved |
| Add minimum skill length constraint (ablation) | DDO · Grid world demonstrations | F1: 0.32 -> 0.35 | Improved |
| Add minimum skill length constraint (ablation) | VTA · Grid world demonstrations | F1: 0.32 -> 0.36 | Improved |
| Add minimum skill length constraint (ablation) | VTA and DDO · Grid world demonstrations | Boundary quality | Mixed |
| Use LOVE skills for downstream exploration | RL agents learning new grid-world tasks · Grid world, sparse reward and N_pick = 5 | Return | Improved |
| Use DDO skills with minimum skill length (ablation) | RL agents learning new grid-world tasks · Grid world, dense reward and N_pick = 5 | Timesteps to solve: over 8× more timesteps | Hurt |
| Use behavior-cloned demonstrations | Low-level policy · Grid world, N_pick = 3 with dense rewards | Return | Improved |
| Use behavior-cloned demonstrations | Low-level policy · Other grid-world settings | Return | No clear effect |
| Use VTA skills for downstream learning | RL agents learning new tasks · Grid world | Return | No clear effect |
| Set compression weight λ to different fixed values (ablation) | LOVE · Simple Colors and Conditional Colors | Boundary F1: Simple: $.67\pm.27$ to $.95\pm.06$; Cond.: $.82 \pm .08$ to $.92 \pm .02$ | Mixed |
| Tune compression weight λ with dual gradient descent (ablation) | LOVE · Simple Colors and Conditional Colors | Boundary F1: Simple: $.99\pm.00$; Cond.: $.99 \pm .00$ | Improved |
| Enforce minimum skill duration during training (ablation) | LOVE · Skill optimization | Optimization stability | Improved |
Closest papers by meaning
- Skills Regularized Task Decomposition for Multi-task Offline Reinforcement Learning (NeurIPS 2022 · similarity 0.88)
- Efficient Skill Discovery via Regret-Aware Optimization (ICML 2025 · similarity 0.87)
- Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · similarity 0.87)
- Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning (ICLR 2024 · similarity 0.86)
- Leveraging Skills from Unlabeled Prior Data for Efficient Online Exploration (ICML 2025 · similarity 0.86)
- Goal-Oriented Skill Abstraction for Offline Multi-Task Reinforcement Learning (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.
