Pedro Rodriguez is a 3rd year PhD candidate in Computer Science at the University of Maryland at College Park advised by Jordan Boyd-Graber. His PhD research is centered on developing algorithms using machine learning and deep learning for natural language processing. His interests include developing question answering systems that automatically answer factoid questions posed by humans in a natural language, finding ways to better understand and interpret machine learning models, and in some cases using those methods to expose the shallow learning that many machine learning models exhibit.
Pedro earned his Bachelor's Degree in Computer Science from the University of California at Berkeley in December 2014. In addition to his PhD research experience, he has worked as a data scientist or research scientist at Riot Games, Zillow Group/Trulia, Oracle, UC Berkeley's AMPLab, UC Berkeley's Astronomy Department, and Boise State University's Cryosphere Group. Pedro was the team captain and a founding member of the Colorado Data Science Team. Combined, Pedro has over 4 years of experience in data science and research science. In his free time he is a volunteer ski patrol, certified avalanche safety instructor, and an avid skier. As an avalanche safety instructor he taught hundreds of students across Idaho, Oregon, California, Colorado, Argentina, and Chile skills to safely enjoy backcountry skiing and riding.
The primary focus of my dissertation research has been in developing an AI (QANTA) that can play Quiz Bowl, a popular trivia game amongst school-aged and college students. This research is funded by a National Science Foundation (NSF) grant with support from an Amazon Web Services (AWS) research grant. You can see our system defeat a strong team of human players 260-215 during our exhibition match at the 2017 Quiz Bowl high school national championships. Note that we did not play against high school students, the opposing team were comprised of helpers at the tournament, many of which had themselves been previous champions in their high school or college days.
A secondary focus of my dissertation research has grown from the intersection of two sources: my advisor's interest in making machine learning models more explainable and interpretable, and my own observation that many models appear to exhibit behaviors attributable to shallow understanding when inspected carefully. My interest is in devising methods that expose these undesirable behaviors, and then attempting to fix them.