Research map / Reinforcement learning / Visual and representation RL
Shaping Belief States with Generative Environment Models for RL
Karol Gregor, Danilo Jimenez Rezende, Frederic Besse, Yan Wu, Hamza Merzic, Aaron van den Oord
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 |
|---|---|---|---|
| Increase overshoot length (ablation) | SimCore agents (all losses) · Random City | position and map decoding accuracy | Improved |
| Use generative prediction loss instead of contrastive for map decoding (ablation) | SimCore · Random City | map decoding MSE: 160 -> 117 | Improved |
| Use contrastive loss instead of generative loss for position decoding (ablation) | SimCore · Random City | position decoding accuracy | Improved |
| Use contrastive loss for mapping (ablation) | Contrastive SimCore · Random City | map decoding | Hurt |
| Add Kanerva memory to generative model (ablation) | Generative SimCore · Random City | map decoding accuracy | Improved |
| Use Kanerva memory instead of pure LSTM or slot-based memory (ablation) | LSTM belief state · Random City | map decoding | Improved |
| Add generative SimCore model loss to LSTM agent | LSTM IMPALA agent · DeepMind Lab | RL score / data efficiency | Improved |
| Add generative SimCore model loss to RMA agent | RMA agent · DeepMind Lab | RL score / data efficiency | Improved |
| Use longer overshoot lengths (ablation) | SimCore agent · DeepMind Lab | RL performance | No clear effect |
| Use Kanerva memory in the RL setting (ablation) | SimCore agent · DeepMind Lab | RL training stability | Hurt |
| Add SimCore to LSTM agent | LSTM agent · Voxel levels (Food, HighFood, Cliff, Bridge) | RL score | Improved |
| Add SimCore to LSTM agent | LSTM agent · Voxel levels | sensitivity to Adam hyper-parameters and unroll length | Improved |
Closest papers by meaning
- Generative Temporal Models with Spatial Memory for Partially Observed Environments (ICML 2018 · similarity 0.91)
- Consistency Models as a Rich and Efficient Policy Class for Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Imagine Beyond ! Distributionally Robust Autoencoding for State Space Coverage in Online Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL (NeurIPS 2026 · similarity 0.86)
- Forecasting in Offline Reinforcement Learning for Non-stationary Environments (NeurIPS 2025 · similarity 0.86)
- RLZero: Direct Policy Inference from Language Without In-Domain Supervision (NeurIPS 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.
