Treballem amb simulació avançada per millorar la seguretat nuclear

Descobreix com la Càtedra UNESCO de Mètodes Numèrics lidera la innovació de frontera al Sud Global

CIMNE Hosts Hackathon to Enhance Kratos Multiphysics with Machine Learning and Optimization

The International Centre for Numerical Methods in Engineering (CIMNE) is hosting a two-day hackathon focused on advancing the integration between Kratos and machine learning (ML) environments, particularly PyTorch. The event, titled “Enhancing Kratos Multiphysics with ML & Optimization,” takes place on June 19-20, 2025, at CIMNE’s Campus Nord UPC facilities in Barcelona.

Organized by the Kratos Multiphysics research group—part of CIMNE’s Large Scale Multiphysics Computations research cluster—the hackathon aims to maximize interoperability between the Kratos framework and machine learning environments. The primary objective is to enable the use of Kratos routines within ML optimization loops, combining forward differentiation with alternative techniques for derivative calculations based on available Jacobians.

The event brings together experts and contributors with working knowledge of Finite Element Methods (FEM) and machine learning, specifically those familiar with the Kratos/PyTorch Python interface. Participation is limited to 20 attendees, all expected to be active contributors to the Kratos GitHub repository.

Comprehensive Two-Day Programme

The hackathon follows a structured approach designed to foster collaboration and practical implementation. Day one focuses on setting the stage and ideation, beginning with an overview of use cases in optimization and Reduced Order Model (ROM) training. Expert presentations will address key challenges and past work, followed by discussions on workflow design, matrix and vector views in Python, and strategies for accessing FEM assembly and Jacobians.

Day two prioritizes prototyping and presentations, with dedicated hands-on hacking sessions allowing teams to develop, debug, and test their solutions. The event concludes with project presentations and feedback sessions, followed by discussions on future collaborations and next steps.

The hackathon targets both the identification of bottlenecks towards ML-Kratos integration and the improvement of existing documentation, addressing critical needs in the computational engineering community.

About Kratos Multiphysics

Kratos Multiphysics is a framework for creating parallel, multidisciplinary simulation software designed for modularity, extensibility, and high performance. Written in C++ with a robust Python interface, Kratos operates under the BSD license, making it suitable for both academic and commercial applications.

The Kratos Multiphysics group at CIMNE develops global-purpose research code integrating state-of-the-art capabilities across multiple fields, with the explicit goal of enabling complex multiphysics problem simulation. The group focuses on exploiting High Performance Computing capabilities for realistic engineering problem simulation through new solution technologies and model integration from different areas.

Afegeix-t’ho al calendari:

Comparteix: