Research map / Reinforcement learning / Visual and representation RL
Temporally Abstract Partial Models
Khimya Khetarpal, Zafarali Ahmed, Gheorghe Comanici, Doina Precup
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 |
|---|---|---|---|
| collect data and mask model-learning loss with the Pickup+Drop@Goal affordance set instead of all options (ablation) | partial option model learning · Taxi | success rate / sample efficiency: 37,500 -> 4,000 state-option pairs | Improved |
| Pickup+Drop (25,000 pairs) instead of Everything (37,500 pairs) during model learning (ablation) | partial option model learning · Taxi | sample efficiency: 37,500 -> 25,000 | No clear effect |
| use any affordance set during model learning allows a higher learning rate (ablation) | partial option model learning · Taxi | training stability (divergence at high learning rate) | Improved |
| apply affordance sets only during planning with the same learned model (ablation) | value iteration over options · Taxi | success rate | No clear effect |
| use the restrictive affordance set in both model learning and planning (ablation) | partial option model · Taxi | sample efficiency | Improved |
| learn affordances online with a classifier instead of hand-designed Pickup+Drop@Goal | partial option model · Taxi | affordance set size and sample efficiency: 4,000 -> 2,000 | Improved |
| increase affordance threshold k so the affordance set shrinks (ablation) | learned affordance classifier · Taxi | sample efficiency | Improved |
| threshold k = 0 (all state-options affordable) (ablation) | learned affordance classifier · Taxi | affordance set size | No clear effect |
Closest papers by meaning
- OptionZero: Planning with Learned Options (ICLR 2025 · similarity 0.87)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.86)
- Reward-Respecting Subtasks for Model-Based Reinforcement Learning (Abstract Reprint) (AAAI 2024 · similarity 0.86)
- Option-aware Temporally Abstracted Value for Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Interpreting Emergent Planning in Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.85)
- On Shallow Planning Under Partial Observability (AAAI 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.
