Accelerated Machine Learning for High Energy Physics

The amount of data produced from High Energy Physics experiments is at a scale which prevents a researcher from thoroughly exploring it using traditional data analysis techniques and this data generation will only grow. Researchers are utilizing machine learning (ML) techniques to increase their ability to analyze these massive datasets. A High Energy Physics researcher now must gain a significant understanding of a machine learning to be able to conduct their analysis. In addition, it can be very easy to build ML tools which introduce more problems than they solve if the developers are not careful. The CFD Research Team, comprised of CFD Research Corporation machine learning experts and Dr. Andreas Jung of the Purdue University High Energy Physics Group, proposes developing an advanced machine learning tool specifically designed for the Department of Energy's high-energy physics (HEP) community. Our proposed tool will provide a user-friendly interface that enables HEP researchers to define their datasets, specify the type of problem they wish to solve, and generate a model with minimal machine learning knowledge. This allows the HEP researchers to focus on the technical problem at hand instead of minutia of machine learning techniques. The Phase I effort will include securing data sets for analysis, definition of search parameters and optimization criteria, and the development of a proof-of-concept tool. The proof-of-concept tool will be targeted at real world problems faced by HEP researchers within the Jung Research Group. It will be tested by the intended end-users to ensure it is well suited for their use-case. Follow-on efforts in Phase II will work to expand the captured use cases of the tool. Additionally, the team will work to ensure the tool is easily deployable across many user facilities and with distributed computing resources. Phase III Plans While the proposed tools will be developed for HEP research, the generalized approach the team is using for the tool will facilitate the commercialization of the framework to non-HEP and non-government customers. Those who desire to leverage machine learning’s capabilities far exceed the number of experts in the field.