Author

Date of Award

2026

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Committee Chair

Joshua Booth

Committee Member

Tathagata Mukherjee

Committee Member

Summer Atkins

Research Advisor

Joshua Booth

Subject(s)

High performance computing, Machine learning, Neural networks (Computer science)

Abstract

Due to rise of machine learning workloads, high performance computing platforms have increased GPU resources. Consequently, use of surrogate models, to utilize GPU compute and speedup scientific applications, is on the rise. Most of the time, the whole program is replaced with a surrogate model. In this work a shock simulation program (LULESH) was taken and three functions of varying complexity were replaced with small sized neural network as surrogate models. The speedup and error in output of whole program was analyzed. We found that for the smallest function, trained model lead to speedup of overall program by 40%, maintained acceptable accuracy and generalized to a bigger domain size. Larger functions however, required bigger model with slower execution time and did not generalize. The implication is that small surrogate models applied to small code segments also provide good potential for optimizing performance.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.