Research map / Reinforcement learning / Visual and representation RL
Combined Reinforcement Learning via Abstract Representations
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 |
|---|---|---|---|
| Add the representation loss L_d to the transition-model objective (ablation) | CRAR · Single labyrinth | abstract-state representation diversity | Helped |
| Replace the representation loss L_d with an auto-encoder reconstruction loss (ablation) | CRAR · Single labyrinth | abstract-state representation diversity | Hurt |
| Set the successive-state resampling proportion β to zero (ablation) | CRAR · Single labyrinth | maze representation quality | No clear effect |
| Increase successive-state resampling in the entropy loss (ablation) | CRAR · Single labyrinth | naturalness of maze representation | Hurt |
| Use a learning rate that is too large (ablation) | CRAR · Single labyrinth | transition-model accuracy | Hurt |
| Add planning at depth D=1, 3, or 6 | CRAR · Distribution of labyrinths with limited off-policy data | score | Helped |
| Remove the model-free component (ablation) | CRAR · Distribution of labyrinths with limited off-policy data | score | Hurt |
| Train a supervised encoder mapping negative images to the same abstract representations as positive images | CRAR · Distribution of labyrinths, negative-image transfer | transfer performance | Helped |
| Remove the model-free, reward, and discount losses in a reward-free labyrinth (ablation) | CRAR · Single labyrinth without rewards | abstract-state representation | No clear effect |
Closest papers by meaning
- Learning Action Representations for Reinforcement Learning (ICML 2019 · similarity 0.89)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.88)
- Learning Generalizable and Composable Abstractions for Transfer in Reinforcement Learning (AAAI 2024 · similarity 0.88)
- Efficient Exploration and Discriminative World Model Learning with an Object-Centric Abstraction (ICLR 2025 · similarity 0.88)
- Reconciling Spatial and Temporal Abstractions for Goal Representation (ICLR 2024 · similarity 0.87)
- Interpreting Emergent Planning in Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.86)
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.