Research map / Reinforcement learning / Visual and representation RL
Influence-Augmented Local Simulators: a Scalable Solution for Fast Deep RL in Large Networked Systems
Miguel Suau, Jinke He, Matthijs T. J. Spaan, Frans Oliehoek
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 |
|---|---|---|---|
| replace the global simulator with an influence-augmented local simulator and pretrained influence predictor | PPO traffic-control training · Traffic control, intersection 1 | training time to match GS-trained performance: about 1/3 of the total training time | Improved |
| replace the global simulator with an influence-augmented local simulator and pretrained influence predictor | PPO warehouse-commissioning training · Warehouse commissioning | training time to reach GS-trained performance: about 1/3 of the total training time | Improved |
| use an untrained influence predictor instead of a predictor trained on global-simulator data (ablation) | IALS · Traffic control | performance on GS | Hurt |
| use an untrained influence predictor instead of a predictor trained on global-simulator data (ablation) | IALS · Warehouse commissioning | performance on GS | Mixed |
| match the influence predictor's history length to the agent's finite memory (ablation) | warehouse IALS and memory-limited agents · Warehouse, items disappear after 8 timesteps | agent performance | Improved |
| use a history-dependent influence predictor rather than a current-observation-only predictor (ablation) | warehouse IALS with memory agents · Warehouse, items disappear after 8 timesteps | agent performance | Improved |
| replace the learned influence predictor with fixed influence probability 0.1 (ablation) | traffic IALS · Traffic control, intersection 1 | performance on GS | No clear effect |
| replace the learned influence predictor with fixed influence probability 0.1 (ablation) | traffic IALS · Traffic control, intersection 2 | performance on GS | No clear effect |
| replace the learned influence predictor with fixed influence probability 0.5 (ablation) | traffic IALS · Traffic control | performance on GS | Hurt |
| replace the learned influence predictor with estimated fixed marginal probabilities (ablation) | warehouse IALS · Warehouse commissioning | performance on GS | Mixed |
Closest papers by meaning
- Distributed Influence-Augmented Local Simulators for Parallel MARL in Large Networked Systems (NeurIPS 2022 · similarity 0.92)
- SRL: Scaling Distributed Reinforcement Learning to Over Ten Thousand Cores (ICLR 2024 · similarity 0.88)
- The Power of Resets in Online Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- A Snapshot of Influence: A Local Data Attribution Framework for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Overcoming the Sim-to-Real Gap: Leveraging Simulation to Learn to Explore for Real-World RL (NeurIPS 2024 · similarity 0.86)
- Counteractive RL: Rethinking Core Principles for Efficient and Scalable Deep Reinforcement Learning (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.
