The GRaM 2026 Competition and Warped-IFW Dataset

Julian Suk, Gavin Seegoolam, Alison Pouplin, Paul Tiwald, Ivan Bioli, Vedant Bonde, Rémi Bourgerie, Oscar Breiner, Thomas Capelle, Huaguan Chen, Alex Colagrande, Aakashnag Davuluri, Vinay Edula, Vivekananda Edula, Bernhard Einberger, Massimiliano Ghiotto, Francesca Maria Greco, Ankit Grover, Justin Hodges, Theofanis Ifaistos, Sahib Julka, Anthony Kalaydjian, Samet Kocbay, Aakash Kotha, Maximilian Leutschafft, Ning Lin, Morgan McGuire, Vlad Medvedev, Mikel Mendibe, Deepthi Ravipati, Jorge Sarrato-Alós, Harshit Singh, Rajeev Kumar Singh, Joshua Stiller, Vihan Tiwari, S. Viswanathan, Luis J. Walter, Andy Zhang
Proceedings of GRaM: the Second Edition of the Workshop on Geometry-grounded Representation Learning and Generative Modeling, PMLR 326:533-602, 2026.

Abstract

Computational fluid dynamics is central to aerodynamic design. Yet, simulating transient and potentially turbulent flow around a new geometry is very expensive. There are increasing efforts to use machine learning models to reduce the cost of simulation. Given the growing use of geometric inductive biases in machine learning, we wanted to know whether architectural choices meaningfully affect results, and whether a competitive geometric solution could emerge. We turned this question into a competition hosted at the GRaM workshop. With BeyondMath, we release Warped-IFW – 181 warped variants of the Imperial Front Wing under detached-eddy flow – and pose one task: given an initial velocity window, predict its continuation. Out of 22 submissions, the top five are all voxel-based. In a post-competition analysis, we find that what actually lowers the error is spatial resolution, not parameter count or architecture family: graph and latent models converge to a plateau that only a finer grid breaks through. Additional details regarding the competition can be found on the competition website.

Cite this Paper

BibTeX
@InProceedings{pmlr-v326-suk26a, title = {The {GRaM} 2026 Competition and {Warped-IFW} Dataset}, author = {Suk, Julian and Seegoolam, Gavin and Pouplin, Alison and Tiwald, Paul and Bioli, Ivan and Bonde, Vedant and Bourgerie, R{\'e}mi and Breiner, Oscar and Capelle, Thomas and Chen, Huaguan and Colagrande, Alex and Davuluri, Aakashnag and Edula, Vinay and Edula, Vivekananda and Einberger, Bernhard and Ghiotto, Massimiliano and Greco, Francesca Maria and Grover, Ankit and Hodges, Justin and Ifaistos, Theofanis and Julka, Sahib and Kalaydjian, Anthony and Kocbay, Samet and Kotha, Aakash and Leutschafft, Maximilian and Lin, Ning and McGuire, Morgan and Medvedev, Vlad and Mendibe, Mikel and Ravipati, Deepthi and Sarrato-Al{\'o}s, Jorge and Singh, Harshit and Singh, Rajeev Kumar and Stiller, Joshua and Tiwari, Vihan and Viswanathan, S. and Walter, Luis J. and Zhang, Andy}, booktitle = {Proceedings of GRaM: the Second Edition of the Workshop on Geometry-grounded Representation Learning and Generative Modeling}, pages = {533--602}, year = {2026}, editor = {Pouplin, Alison and Vadgama, Sharvaree and Bekkers, Erik and Kaba, Sékou-Oumar and Lawrence, Hannah and Lecha, Manuel and Baker, Elizabeth and Suk, Julian and Walters, Robin and Tomczak, Jakub and Jegelka, Stefanie}, volume = {326}, series = {Proceedings of Machine Learning Research}, month = {26 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v326/main/assets/suk26a/suk26a.pdf}, url = {https://proceedings.mlr.press/v326/suk26a.html}, abstract = {Computational fluid dynamics is central to aerodynamic design. Yet, simulating transient and potentially turbulent flow around a new geometry is very expensive. There are increasing efforts to use machine learning models to reduce the cost of simulation. Given the growing use of geometric inductive biases in machine learning, we wanted to know whether architectural choices meaningfully affect results, and whether a competitive geometric solution could emerge. We turned this question into a competition hosted at the GRaM workshop. With BeyondMath, we release Warped-IFW – 181 warped variants of the Imperial Front Wing under detached-eddy flow – and pose one task: given an initial velocity window, predict its continuation. Out of 22 submissions, the top five are all voxel-based. In a post-competition analysis, we find that what actually lowers the error is spatial resolution, not parameter count or architecture family: graph and latent models converge to a plateau that only a finer grid breaks through. Additional details regarding the competition can be found on the competition website.} }
Endnote
%0 Conference Paper %T The GRaM 2026 Competition and Warped-IFW Dataset %A Julian Suk %A Gavin Seegoolam %A Alison Pouplin %A Paul Tiwald %A Ivan Bioli %A Vedant Bonde %A Rémi Bourgerie %A Oscar Breiner %A Thomas Capelle %A Huaguan Chen %A Alex Colagrande %A Aakashnag Davuluri %A Vinay Edula %A Vivekananda Edula %A Bernhard Einberger %A Massimiliano Ghiotto %A Francesca Maria Greco %A Ankit Grover %A Justin Hodges %A Theofanis Ifaistos %A Sahib Julka %A Anthony Kalaydjian %A Samet Kocbay %A Aakash Kotha %A Maximilian Leutschafft %A Ning Lin %A Morgan McGuire %A Vlad Medvedev %A Mikel Mendibe %A Deepthi Ravipati %A Jorge Sarrato-Alós %A Harshit Singh %A Rajeev Kumar Singh %A Joshua Stiller %A Vihan Tiwari %A S. Viswanathan %A Luis J. Walter %A Andy Zhang %B Proceedings of GRaM: the Second Edition of the Workshop on Geometry-grounded Representation Learning and Generative Modeling %C Proceedings of Machine Learning Research %D 2026 %E Alison Pouplin %E Sharvaree Vadgama %E Erik Bekkers %E Sékou-Oumar Kaba %E Hannah Lawrence %E Manuel Lecha %E Elizabeth Baker %E Julian Suk %E Robin Walters %E Jakub Tomczak %E Stefanie Jegelka %F pmlr-v326-suk26a %I PMLR %P 533--602 %U https://proceedings.mlr.press/v326/suk26a.html %V 326 %X Computational fluid dynamics is central to aerodynamic design. Yet, simulating transient and potentially turbulent flow around a new geometry is very expensive. There are increasing efforts to use machine learning models to reduce the cost of simulation. Given the growing use of geometric inductive biases in machine learning, we wanted to know whether architectural choices meaningfully affect results, and whether a competitive geometric solution could emerge. We turned this question into a competition hosted at the GRaM workshop. With BeyondMath, we release Warped-IFW – 181 warped variants of the Imperial Front Wing under detached-eddy flow – and pose one task: given an initial velocity window, predict its continuation. Out of 22 submissions, the top five are all voxel-based. In a post-competition analysis, we find that what actually lowers the error is spatial resolution, not parameter count or architecture family: graph and latent models converge to a plateau that only a finer grid breaks through. Additional details regarding the competition can be found on the competition website.
APA
Suk, J., Seegoolam, G., Pouplin, A., Tiwald, P., Bioli, I., Bonde, V., Bourgerie, R., Breiner, O., Capelle, T., Chen, H., Colagrande, A., Davuluri, A., Edula, V., Edula, V., Einberger, B., Ghiotto, M., Greco, F.M., Grover, A., Hodges, J., Ifaistos, T., Julka, S., Kalaydjian, A., Kocbay, S., Kotha, A., Leutschafft, M., Lin, N., McGuire, M., Medvedev, V., Mendibe, M., Ravipati, D., Sarrato-Alós, J., Singh, H., Singh, R.K., Stiller, J., Tiwari, V., Viswanathan, S., Walter, L.J. & Zhang, A.. (2026). The GRaM 2026 Competition and Warped-IFW Dataset. Proceedings of GRaM: the Second Edition of the Workshop on Geometry-grounded Representation Learning and Generative Modeling, in Proceedings of Machine Learning Research 326:533-602 Available from https://proceedings.mlr.press/v326/suk26a.html.

Related Material