Co-design Center for Exascale Machine Learning Technologies (ExaLearn)

Abstract

Rapid growth in data, computational methods, and computing power is driving a remarkable revolution in what variously is termed machine learning (ML), statistical learning, computational learning, and artificial intelligence. In addition to highly visible successes in machine-based natural language translation, playing the game Go, and self-driving cars, these new technologies also have profound implications for computational and experimental science and engineering, as well as for the exascale computing systems that the Department of Energy (DOE) is developing to support those disciplines. Not only do these learning technologies open up exciting opportunities for scientific discovery on exascale systems, they also appear poised to have important implications for the design and use of exascale computers themselves, including high-performance computing (HPC) for ML and ML for HPC. The overarching goal of the ExaLearn co-design project is to provide exascale ML software for use by Exascale Computing Project (ECP) applications, other ECP co-design centers, and DOE experimental facilities and leadership class computing facilities.

Autonomous x-ray scattering

“…The agent's goal is to select actions that maximize an encoded scientific goal. This reinforcement learning approach has been highly successful in game-playing [102,103], for instance, and shows promise for rapid decision-making in experimental contexts [16,41]. An alternative approach is to treat AE as an optimization problem in a multi-dimensional parameter space, which is defined as a set of signals-of-interest that vary over the range of control parameters.…”

“…Alternatively, other decision-making modules should be considered when very high-speed decision making is required. Reinforcement learning methods that exploit neural network architectures can achieve the higher speed required for certain experiments [16].…”

“…Synchrotron beamtime-the focus of this review-is one example of a high-value resource that should be used as efficiently and completely as possible. Cutting-edge artificial intelligence (AI) and machine-learning (ML) methods are an attractive option for handling this complexity, and have been deployed to positive effect to automate many scientific tasks, including streamlining data analysis or simulation [2,3,[11][12][13][14][15][16][17].…”