At the ALS we, as members of the Instrumentation-Thrust Area, are in a unique position to lead the development of new experimental and computational approaches in a synchrotron environment. Recent breakthroughs in machine learning have had a tremendous impact across industry and science domains, largely due to the readily available large labeled data sets and new developments in more effective network architectures.

However, there are several challenges associated with capitalizing on these ideas in the context of scientific machine learning. Even if the data sets created at the ALS are by their nature large (e.g. 4D STXM and ptychography, nanoARPES, XPCS, ... ), they usually only represent a subset of materials and parameter space. Extrapolation from there to other materials and properties can be very challenging.

The combination of the ALS, the Molecular Foundry, NERSC, and our collaborating colleagues at CAMERA, MSD, CSD, UCB, and all other academic and industrial research activities of the Bay Area provide a unique environment to tackle this and other challenges of scientific machine learning.

Read the most recent AI-ML DOE report below