Adam Harrison, right, with his computer engineering supervisor Dileepan Joseph, has argued for paradigm change in computing to help computers convert 2-D images into 3-D shapes and textures.
(Edmonton) It hardly qualifies as light weekend reading but Adam Harrison's PhD thesis, entitled Numeric Tensor Framework: Toward a New Paradigm in Technical Computing, at least delivers more than the typical computer engineering thesis.
"It asks questions other people don't usually ask," says computer engineering professor Dileepan Joseph, who supervised Harrison's PhD research. "It has software, data analysis, and applications, but it also has rhetoric. It makes an argument founded in philosophy and that alone is unusual-that makes it stand out from other theses I have read."
Harrison's thesis won the Spring 2016 George B. Walker Award for Best Doctoral Thesis in the Department of Electrical and Computer Engineering. The external examiner, a professor of computing, wrote that the thesis was "definitely the most eloquently expressed doctoral dissertation" of more than 200 such theses he had reviewed over his career.
"I never expected to hear a reviewer say anything like that," said Harrison, who is conducting post-doctoral research on medical imaging analysis in Bethesda, Maryland, for the National Institutes of Health (NIH). "I was very flattered by that remark."
The thesis builds on mathematical notation developed by Einstein a century ago and arguments on paradigms and scientific progress, made by the late American physicist and philosopher Thomas S. Kuhn. Harrison builds off of existing arguments that computer simulation and computation form a third pillar of scientific progress, alongside experimentation and theory. Within this third pillar, he argues that a paradigm, which he calls the matrix/vector paradigm, applies to technical computing.
According to Kuhn, all paradigms have limits and paradigm change may be required to resolve anomalies, which are significant limitations. In one motivating application of Harrison's research, namely the conversion of 2-D images into 3-D shapes and textures, Harrison ran up against the limits of the matrix/vector paradigm. Recognizing a need for new algebra and software to model and compute solutions to such problems, Harrison and Joseph developed the numeric tensor framework, which they recently published in the Journal of Computational Science.
"We're interested in image processing and computer vision," said Harrison. "These fields fundamentally challenge the matrix/vector paradigm because it can't handle multi-dimensional data and the mathematical mappings that you need to perform.
"In my research, I discovered that there's a whole slew of researchers having similar problems and coming up with their own solutions to tackle these common anomalies, sometimes by shoehorning the matrix/vector paradigm."
For a long time, researchers from many fields have been encountering and wrestling with these anomalies within their own areas of research but Harrison recognized the similarities across disciplines and the role of a paradigm in obstructing progress, says Joseph.
"There is a cross-disciplinary need for something beyond the matrix/vector paradigm, and this thesis makes an important contribution to that effort."
Harrison returned to Edmonton in early June for his convocation ceremony but was keen to get back to his work in the U.S. He's working on computer-aided diagnostics with medical professionals at the NIH Clinical Center, the largest research hospital in the world.
"I'm working right in the clinic every day, alongside doctors and patients," he said. The work is challenging and Harrison is excited about the future of medical diagnostics and imaging.
"We're starting to apply new technologies, like deep-learning, but we've just scratched the surface so far. Of course, I'm also thinking of ways in which my PhD work could help."