Publication Date

8-2015

Date of Final Oral Examination (Defense)

6-29-2015

Type of Culminating Activity

Thesis

Degree Title

Master of Science in Computer Science

Department

Computer Science

Supervisory Committee Chair

Amit Jain, Ph.D.

Supervisory Committee Member

Hans-Peter Marshall, Ph.D.

Supervisory Committee Member

Timothy Andersen, Ph.D.

Abstract

To fully understand the complex interactions of various phenomena in the natural world, scientific disciplines such as geology and seismology increasingly rely upon analyzing large amounts of observations. However, data collection is growing at a faster rate than what is currently possible to analyze through traditional approaches. These datasets, supplied by the increasing use of sensors and remote sensing, require specialized computer programs to effectively analyze complex and expansive volumes of data.

Elaborating on existing geophysical data processing approaches for infrasound data collected from an avalanche-prone area, this project proposes new techniques for processing large geophysical datasets. These improved techniques take advantage of Graphical Processing Units (GPU) to accelerate floating-point, computationally expensive operations. Additionally, they allow for dividing the workload among nodes in a High Performance Computing cluster yielding a performance speedup of 3.5 times for every additional node added. Finally, a machine learning approach was used to classify events found in the processed data, which demonstrates the potential of automatic real-time avalanche detection.

Applications with characteristics similar to infrasound processing are common throughout the earth-sciences, and this work exemplifies the potential of these techniques to an array of science fields. New algorithms for efficient data processing like those presented in this work are fundamental to analyzing large geophysical datasets, as well as to improving the accuracy of computer models across many disciplines.

Download

Share

COinS