Thomas Lux does not hesitate when it comes to setting long-term goals.
“After graduation I would like to work somewhere that allows me to devote my time to pursuing research in artificial general intelligence,” he said. “I can easily see myself at an industry/government lab, in academia, or in a small startup. I will be happy as long as I get to contribute to the creation of super-human intelligent algorithms that can benefit people in society.”
Lux, a computer science student in the Discovery Analytics Center and a research trainee in the National Science Foundation-sponsored Urban Computing certificate program, an interdisciplinary program administered through DAC, would also like to use his data analytics skills to collaborate and contribute to his fiancé’s work in neuropsychological assessment. (She is pursuing a Ph.D. in clinical neuropsychology at Saint Louis University.)
In choosing a Ph.D. program, he was particularly interested in working with faculty like his advisor, Layne Watson, who have strong backgrounds in mathematics and optimization, and whose research is grounded in practical applied problems.
“I think the best way to solve problems is by learning to combine existing theory with real-world constraints in order to develop new theory tailored for specific applications,” Lux said.
Lux’s research focuses on applied approximation, numerical analysis, and nonparametric statistics. He is part of the VarSys team that creates models to help understand and manage computational performance variability across the computer system stack.
Petascale — and inevitably exascale — computing comes with many hurdles, he said, including immense losses in performance that result from interactions between parts of computers. The team collects data and builds models that address questions like: How do you configure an operating system to minimize energy consumption? What file and record sizes should be used to minimize read throughput variance? What CPU cache hierarchy is least vulnerable to side channel attacks?
Lux is committed to contributing to this field of study and this summer will be his third consecutive doing research at VarSys.
“Although the systems-oriented application that I work on may sound far off, the underlying mathematical concepts are surprisingly similar. In order to invent intelligent learning algorithms, we must understand the limits and maximize the performance of models we build over data. My theoretical work for systems is laying a foundation of robust, mathematically justified, and explainable algorithms for creating learning machines,” Lux said.
Lux’s research as first author has been included in a number of conference proceedings. Among them are: Nonparametric Distribution Models for Predicting and Managing Computational Performance Variability, IEEE SoutheastCon 2018; Predictive Modeling of I/O Characteristics in High Performance Computing Systems, Association of Computing Machinery High Performance Computing Symposium, April 2018; and Novel Meshes for Multivariate Interpolation and Approximation, 2018 Association of Computing Machinery Southeast Conference.
He has also coauthored a number of papers.
When not busy working on his research, Lux enjoys active sports — soccer, frisbee golf, and racquetball — and hiking the mountains of southwest Virginia. Formally trained in jazz percussion and drums, he said he loves music but now mostly plays piano and guitar.
Lux’s projected graduation is May 2020.