Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates.
Improving Model-Based Reinforcement Learning by Converging to Flatter Minima
Authors: Shrinivas Ramasubramanian, Benjamin Freed, Alexandre Capone, Jeff G. Schneider
NeurIPS 2025 | Venue PDF | LLM Run Details | Input Tokens: 26,247 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 3,069 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, SAM reduces measured sharpness and value-prediction error and improves returns across Humanoid Bench, Atari-100k, and high-Do F Deep Mind Control tasks. Augmenting existing MBRL algorithms with SAM increases mean return, with especially large gains in settings with high dimensional state action spaces. |
| Researcher Affiliation | Academia | Robotics Institute, Carnegie Mellon University; Pittsburgh, PA 15213 EMAIL |
| Pseudocode | No | The paper describes algorithms and mathematical formulations but does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block. |
| Open Source Code | Yes | 1Code: https://github.com/autonlab/MBRL-flat-minima.git |
| Open Datasets | Yes | Empirically, SAM reduces measured sharpness and value-prediction error and improves returns across Humanoid Bench, Atari-100k, and high-Do F Deep Mind Control tasks. |
| Dataset Splits | No | The paper discusses training durations like '2M environment steps' and '100k environment steps' and 'running four random seeds' which are common in RL benchmarks, but it does not specify explicit training/validation/test dataset splits in terms of percentages, sample counts, or specific predefined files for static datasets. |
| Hardware Specification | No | We would like to thank Pittsburgh Supercomputer Center for providing access to their HPC resources. However, the specific types (e.g., GPU models, CPU types) of these resources are not detailed. |
| Software Dependencies | No | The paper mentions using existing algorithms and frameworks (e.g., TD-MPC2, TWISTER) and implicitly relies on standard ML libraries, but it does not provide specific version numbers for any software dependencies like Python, PyTorch, or other libraries. |
| Experiment Setup | Yes | Baseline hyper-parameters follow the original papers; SAM introduces a single radius parameter ρ App. Tab. 4. Unless otherwise noted, all hyper-parameters follow the TD-MPC2 defaults, and the SAM radius ρ is selected via a small coarse grid in App. Tab. 5. |