PhD student Raymond Alexander Yeh has been selected as one of 39 recipients of a 2018 Google PhD Fellowship. Since its inception in 2009, nine other Illinois PhD students have been named Fellows in the highly selective program, but Yeh is the first from ECE. The Google Fellowship offers two years of support, and it includes annual stipend of $34,000, coverage of tuition and fees, and the opportunity to work with a Google research mentor.
"The student nominations we received were exemplary in their quality, but Raymond especially stood out and was endorsed by the research scientists and distinguished engineers within Google who participated in the review," wrote Susie Kim, Google PhD Fellowships.
Yeh is jointly advised by Minh N Do, Mark Hasegawa-Johnson, and Alexander Schwing. All four are affiliated with the Coordinated Science Lab. Do and Hasegawa-Johnson are also affiliated with Beckman Institute.
In his proposal, "Interpretable Learning Based Computer Vision Systems," Yeh aims to build interpretable computer vision systems, focusing on two tasks: textual grounding and image restoration. Since computer vision applications are unique, there isn't likely a single solution for all problems. However, interpretable models can reduce debugging time and lead to systemic improvements.
Textual grounding is highly applicable in VR and AR environments. While it's easy for a person to look at a grocery shelf and select "the small can on the right," it is challenging for current textual grounding algorithms to interpret the phrase, parse the information, and apply it to the visual environment. Yeh's approach cast the problem of textual grounding into a unified framework that permits efficient search over all possible bounding boxes and provides interpretability on the trained model parameters.
Yeh is addressing image restoration, the other task within his proposal, by bringing the image generation models and image prior choice together. Current processes keep these independent, but Yeh proposes to leverage recent advances in Generative Adversarial Networks (GANs).
The Google Fellowship program invites select universities to submit up to two nominations each for the national competition. The Graduate College coordinates the selection process to determine the nominees from Illinois.
"Given the caliber of our programs in the computational sciences and related fields, the campus' Google selection panel has the unenviable job of determining who among the stellar applicants is the most competitive," said Ken Vickery, director of fellowships in the Graduate College. "We are truly proud to see Raymond's work recognized and rewarded with such a tremendous accolade."
Article by Julia Sullivan, ECE ILLINOIS