We are actively developing a suite of machine learning benchmarks for scientific datasets using facility data from across STFC and the wider community. Benchmarking machine learning approaches applied to scientific datasets is of interest for several reasons:
- Benchmarking provides platform for fair comparison of methods, software, and hardware approaches to solving scientific problems with machine learning.
- Benchmarking facilitates the collection of curated datasets which can be used to motivate new approaches to tackling the problem.
- Benchmarking provides a set of reference implementations demonstrating applications of various machine learning approaches which can be used to educate and inspire.
