Research map / Probabilistic and scientific modeling / Dynamical systems
Better Transfer Learning with Inferred Successor Maps
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 Bayesian Successor Representation with inferred reward-context clusters and multiple successor maps instead of a single successor map | Successor representation · Grid-world maze, Experiment I | Steps (10^3): $34.1\pm0.7$ vs. SSR $39.8\pm0.9$ | Helped |
| Use Bayesian Successor Representation with inferred reward-context clusters and multiple successor maps instead of a single successor map | Successor representation · Continuous maze, Experiment III | Total rewards (10^3): $36.1\pm0.5$ vs. SSR $24.1\pm0.3$ | Helped |
| Add a constant reward-weight offset (ablation) | BSR-4 · Puddle-world, Experiment II | Returns (10^3): $13.3\pm0.5$ -> $23.2\pm0.4$ | Helped |
| Add a convolved-reward (CR) exploration offset alongside the constant offset (ablation) | BSR-4 · Puddle-world, Experiment II | Returns (10^3): $23.2\pm0.4$ -> $27.1\pm0.5$ | Helped |
| Add a convolved-reward exploration offset alongside the constant offset (ablation) | SSR · Puddle-world, Experiment II | Returns (10^3): $20.1\pm0.4$ -> $17.9\pm0.8$ | Hurt |
| Update only the sampled map rather than the most likely map (ablation) | BSR-4 · Continuous maze, Experiment III | Total rewards (10^3): $36.1\pm0.5$ -> $32.4\pm0.4$ | Hurt |
| Use more stored maps in GPI (10 instead of 4) (ablation) | GPI · Grid-world maze, Experiment I | Steps to complete episodes | Hurt |
| Use full particle-filtering GSR instead of the approximate BSR inference (ablation) | BSR · Grid-world maze, Experiment I | Steps to complete episodes | No clear effect |
| Use full particle-filtering GSR instead of the approximate BSR inference (ablation) | BSR · Puddle-world, Experiment II | Performance | Hurt |
| Update the CR maps only at episode end rather than during the episode (ablation) | BSR2 · Experiments using BSR2 | Performance | No clear effect |
| Update only the most likely successor map instead of updating all maps (ablation) | BSR · Tabular setting | Performance | No clear effect |
| Update only the sampled successor map instead of updating all maps (ablation) | BSR · Tabular setting | Performance | Hurt |
Closest papers by meaning
- Learning Place Cell Representations and Context-Dependent Remapping (NeurIPS 2024 · similarity 0.88)
- Place Cells as Multi-Scale Position Embeddings: Random Walk Transition Kernels for Path Planning (NeurIPS 2025 · similarity 0.87)
- Predictive auxiliary objectives in deep RL mimic learning in the brain (ICLR 2024 · similarity 0.87)
- Time Makes Space: Emergence of Place Fields in Networks Encoding Temporally Continuous Sensory Experiences (NeurIPS 2024 · similarity 0.87)
- Disentangling Representations through Multi-task Learning (ICLR 2025 · similarity 0.86)
- A Model of Place Field Reorganization During Reward Maximization (ICML 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.