As they are promising to reduce modeling and computational effort in the long run, Machine Learning methods get increasingly into the focus of physics and engineering. They are e. g. used to accelerate atomistic simulations, to represent macroscopic and homogenized material behavior based on microstructure information as a generalized approach and to facilitate multi-physics simulations of structures. A major current challenge is to make these methods robust, efficient and reliable for a broad range of application cases so that they can used without case-specific adaptations that would require detailed knowledge of the use-case and the ML model itself. Improvements in this area will give researchers from neighbored fields the opportunity to use these models independently and counters the still poor availability of public implementations of conventional models.

The workshop shall present examples for current approaches to this challenge, discuss impacts on efficiency and robustness in detail and to serve as an exchange platform between the domains of science, engineering and computer sciences and for best practice solutions.