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..
Bridging Equivariant GNNs and Spherical CNNs for Structured Physical Domains
Authors: Colin Kohler, Purvik Patel, Nathan Vaska, Justin Goodwin, Matthew Jones, Robert Platt, Rajmonda Caceres, Robin Walters
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We perform experiments on various challenging domains, including radar response modeling, aerodynamic drag prediction, and policy learning for manipulation and navigation. We find that G2Sphere outperforms competitive baselines in terms of accuracy and inference time, and we demonstrate that equivariance and Fourier features lead to improved sample efficiency and generalization. |
| Researcher Affiliation | Collaboration | Colin Kohler Northeastern University Boston, MA 02115 EMAIL Purvik Patel Northeastern University Boston, MA 02115 EMAIL Nathan Vaska MIT Lincoln Laboratory Lexington, MA 02421 EMAIL Justin Goodwin MIT Lincoln Laboratory Lexington, MA 02421 EMAIL Matthew C. Jones MIT Lincoln Laboratory Lexington, MA 02421 EMAIL Robert Platt Northeastern University Boston, MA 02115 EMAIL Rajmonda S. Caceres MIT Lincoln Laboratory Lexington, MA 02421 EMAIL Robin Walters Northeastern University Boston, MA 02115 EMAIL |
| Pseudocode | No | The paper describes methods and architectures but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks or figures. |
| Open Source Code | Yes | The source code is available at: https://github.com/Colin Kohler/geometry2sphere. |
| Open Datasets | No | While we are planning on releasing both the code and datasets in this work if accepted, they are subject to an internal review process which must be complete first which will began upon acceptance of this work (if accepted). |
| Dataset Splits | Yes | Each dataset is divided 90/10 between train/validation and test sets. |
| Hardware Specification | Yes | All of our experiments are run on a high-performance cluster using either one or two Nvidia Xeon-g6-volta GPUs with 32GB of memory. Each node has a Intel Xeon Gold 6248 2.4 GHz CPU with 128GB of RAM and 1TB of local storage. |
| Software Dependencies | No | The paper mentions specific optimizers (e.g., Adam optimizer [32]) and architectures (e.g., Equiformer v2 [40]) and refers to implementations from other papers (e.g., Diffusion models [13]), but it does not specify version numbers for general software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | We train using the Adam optimizer [32] with the best learning rate and its decay were chosen to be 5e 4 and 1e 5 respectively. We use a batch size of 4. |