Sponsored by the U.S. Department of Energy (DOE) Advanced Scientific Computing Research (ASCR), the two-part online workshop began on December 2–3, 2020, with a bootcamp including interactive tutorial presentations, panel sessions with application specialists, and extensive Q&A with experts. More than 450 participants attended the workshop. The second part, held January 6–7, 2021, involved intensive brainstorming and writing sessions.

Researchers from the Mathematics and Computer Science (MCS) division at Argonne National Laboratory played a major role in the workshop. Stefan Wild, a senior computational mathematician and deputy director of the MCS division, was co-chair of the workshop and lead of the report subsection on Software and Libraries for Scientific Computing; and senior mathematician Mihai Anitescu was lead of the report subsection on Complexity Analysis. Additional members from the MCS division contributing to the workshop report writing were Wendy Di, Carlo Graziani, Paul Hovland and Ruslan Shaydulin.

The RASC workshop defined randomized algorithms as “those algorithms that employ some form of randomness in internal algorithmic decisions to accelerate time to solution, increase scalability, or improve reliability. Examples include matrix sketching for solving large-scale least squares problems and stochastic gradient descent for training machine learning models.”

While randomized algorithms have long been used in applications such as Internet protocols, cryptography, and signal processing, interest in such algorithms has been accelerated by new results in artificial intelligence (AI). The workshop sought to address the new skills particularly needed to advance AI for Science within DOE/ASCR.

Based on discussions and community input, the workshop participants recommended a research program comprising six interrelated threads:

• Foundational research in the theory of randomized algorithms
• Development of sophisticated algorithms that leverage the theoretical underpinnings in practice
• Deployment in scientific applications, including software, benchmarks, and best practices for specific applications
• Adaptation of randomized algorithms to take advantage of computing hardware, from current architectures to quantum, neuromorphic, and other emerging platforms
• Outreach to a broader community, including experts in statistics, applied probability, signal processing, and emerging hardware
• Workflow standardization, including testbeds and modular frameworks for incorporating new methods and deploying them in new architectures

The full workshop report, with details about the application drivers motivating the research and candidate research directions for enabling randomization in scientific computing, is available on the web.