Notice: 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. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Diversifying Parallel Ergodic Search: A Signature Kernel Evolution Strategy
Authors: Sreevardhan Sirigiri, Christian Hughes, Ian Abraham, Fabio Ramos
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we present results to empirically demonstrate the effectiveness and applicability of the methods discussed above in a variety of simulated and real experiments (and their exact setup can be found in Appendix K). |
| Researcher Affiliation | Collaboration | Sreevardhan Sirigiri The University of Sydney, Australia EMAIL Christian Hughes Yale University, USA EMAIL Ian Abraham Yale University, USA EMAIL Fabio Ramos Nvidia, USA The University of Sydney, Australia EMAIL |
| Pseudocode | Yes | Algorithm 1 Stein Variational CMA ES Algorithm 2 Stein Variational Ergodic Trajectory Opt. Algorithm 3 Stein Variational Ergodic Control |
| Open Source Code | No | No. We extended the code used from the paper [2]. The work in this paper is closed source and therefore we did not open our code. |
| Open Datasets | No | No. The paper describes simulated environments and custom setups for its experiments (e.g., 'We consider the two-dimensional spatial domain S = [0, 100] [0, 100] m, equipped with the uniform measure π' or 'We consider a three-dimensional domain S = [0, 3] [0, 3] [0.5, 1.5]'). There are no explicit links, DOIs, or citations to publicly available datasets used for evaluation. |
| Dataset Splits | No | No. The paper describes simulation-based experiments where trajectories and control sequences are generated. It does not use pre-existing datasets that would require explicit training/test/validation splits. For instance, in Section 7.1.1, it states 'For each prior, we generate N = 6 trajectory samples'. |
| Hardware Specification | Yes | Hardware used All experiments were run on the a computer with: 1. Operating System: Ubuntu 20.04.6 LTS 2. CPU: AMD Ryzen 9 5900HS 3. RAM: 16 GB 4. GPU: NVIDIA Ge Force RTX 3060 5. CUDA Version: 12.2 |
| Software Dependencies | No | No. While CUDA Version 12.2 is mentioned, the paper does not specify version numbers for other key software components or libraries used from the referenced codebases [2, 22, 20] or the 'softdtw_jax' GitHub repository, which are crucial for replication. |
| Experiment Setup | Yes | The experimental results are summarized in Figure 3 and an elaborate statistical analysis can be found in Appendix J (Table 2). All Stein variational gradient updates employ a step size of ϵ = 0.01, and iterations are terminated when x(i) x(i 1) < 1.25 10 3 or maximum of 3000 iterations is reached. Table 5: Full Hyperparameter overview of SV-CMA-ES in the multiscale constrained exploration example |