Matthew Feickert
June 27, 2025
Doing interesting research can be hard, and having to carefully curate a complex software stack of tools by hand or debug why your software environment broke when it worked two days ago can make it even harder. Luckily, we don’t have to make research harder than it needs to be!
Scientific researchers need reproducible software environments for complex applications that can run across heterogeneous computing platforms. We now have modern tools for creating fully reproducible hardware accelerated software environments for machine learning workflows (and other scientific applications that use CUDA) that use high level semantics aimed at researchers!
If this sounds interesting or useful to your research, please apply for the national level 2025 Reproducible Machine Learning Workflows for Scientists Workshop! As part of the URSSI 2025 Early-Career Fellowship program, there will be a three day workshop (August 12-14, 2025) on methods for reproducible scientific software environments for artificial intelligence/machine learning (AI/ML) and hardware accelerated workflows at the University of Wisconsin–Madison.
The workshop will provide a practical introduction to using modern open source tools, like Pixi, to easily create computing environments for scientific and AI/ML workflows that benefit from hardware acceleration across multiple machines and platforms. It will also cover how the modern CUDA software stack — from compilers to development libraries — can now be leveraged through conda packages, allowing for fine grain control of CUDA dependencies. The workshop will use motivating examples of applications of CUDA accelerated Python machine learning libraries, and cover how to deploy these environments to production computing systems with GPUs using Linux container images.
This workshop will not teach machine learning concepts, but will focus on the methodologies and tools to make existing machine learning workflows reproducible. As the workshop will also include afternoon “hands on” work time with the instructor team to apply the day’s lessons to real world research, all workshop participants are strongly encouraged to bring a research problem or idea with them to work on.
The workshop material was designed for a target audience of “early career researchers”, but that should be broadly interpreted. All career stages (from students to faculty and staff) are welcome to apply and participate! Workshop applicants who might benefit the most from this workshop are researchers who:
All of them! You do not need to be a machine learning researcher or engineer to participate in this workshop. We are targeting a diverse participant group from all areas of science and engineering to maximize the workshop impact.
The workshop expects participants to at minimum have basic experience at the level taught at a Software Carpentry workshop with:
No prior experience with machine learning tools and technologies is required, though experience is beneficial. Researchers with interest in general hardware acceleration for computing are encouraged to apply as well.
The workshop is free to participate in and there are no associated fees, thanks to support by the US Research Software Sustainability Institute (URSSI) via grant G-2022-19347 from the Sloan Foundation, and the University of Wisconsin–Madison Data Science Institute.
The workshop is not able to provide travel support, but for a limited number of applicants the workshop will be to cover three nights of lodging at the workshop hotel.