Research map / Reinforcement learning / MDPs and planning
Flowing Through States: Neural ODE Regularization for Reinforcement Learning
Mohamed Ghanem, Bernd Finkbeiner
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 |
|---|---|---|---|
| add FlowReg (Index sampling) to A2C | A2C · Atari Qbert | mean episode reward: 4374.30 -> 8306.05 | Improved |
| add FlowReg (Index sampling) to A2C | A2C · Atari Riverraid | mean episode reward: 1862.27 -> 2946.34 | Improved |
| add FlowReg (Exp-Decay sampling) to A2C | A2C · Atari BeamRider | mean episode reward: 960.66 -> 1593.11 | Improved |
| use Index instead of Exp-Decay time sampling (ablation) | FlowReg · Atari Qbert | mean episode reward: 8306.05 vs 6903.15 | Mixed |
| apply FlowReg every 10 updates instead of every 5 (Exp-Decay) (ablation) | FlowReg · Atari Qbert (Exp-Decay) | mean episode reward: 5286.60 -> 6903.15 | Improved |
| apply FlowReg every 5 updates instead of every 10 (Index) (ablation) | FlowReg · Atari Qbert (Index) | mean episode reward: 6569.51 -> 8306.05 | Improved |
| lower FlowReg frequency to once every 20 updates (ablation) | FlowReg · Atari Qbert | mean episode reward: 4374.30 -> 5985.70 | No clear effect |
| add FlowReg to A2C smoothness of latent paths | A2C · Atari Qbert | latent path length (normalized): 34.39 -> 4.20 | Improved |
| add FlowReg to A2C acceleration energy | A2C · Atari Breakout | acceleration energy (normalized): 31432.59 -> 94.98 | Improved |
| replace TACO with FlowReg | A2C · Atari Breakout | mean episode reward: 11.12 -> 24.03 | Improved |
| add TACO to A2C | A2C · Atari Qbert | mean episode reward: 4374.30 -> 2434.05 | Hurt |
| add FlowReg to PPO | PPO · Minigrid FourRooms and Dynamic-Obstacles | episode reward | Improved |
Closest papers by meaning
- From Ticks to Flows: Dynamics of Neural Reinforcement Learning in Continuous Environments (ICLR 2026 · similarity 0.88)
- Temporal Regularization for Markov Decision Process (NeurIPS 2018 · similarity 0.88)
- ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow (NeurIPS 2026 · similarity 0.87)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.86)
- Normalizing Flows are Capable Models for Continuous Control (NeurIPS 2025 · similarity 0.86)
- Reinforcement Learning Under Latent Dynamics: Toward Statistical and Algorithmic Modularity (NeurIPS 2024 · 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.
