Michael Pazzani attended some of the first machine learning workshops, surrounded by a handful of other pioneers who had high hopes for this incongruous topic, back in the mid 1980s. Since those early days, he has managed budgets of a quarter billion dollars to study this discipline and met with the Pentagon to advise the Secretary of Defense.
Last November, he joined USC’s Information Sciences Institute (ISI) after reaching out to its executive director, Craig Knoblock, another machine learning pioneer. “I told him that ISI was the only place in L.A. where I wanted a position,” recalls Pazzani. Ironically, this was not his first time applying. “I interviewed at ISI in 1984 when I was at UCLA as a grad student. At that point ISI – affiliated with the competitor USC – had difficulties supporting UCLA grad students, so it made sense for me to go elsewhere.”
Knoblock did not need to think twice about this collaboration. “Michael brings years of experience. I expect him to both help mentor young ISI researchers in launching their own research careers and lead new initiatives, such as building closer ties with the USC Keck Medical School.” Pazzani is also well versed in the art of grant writing after 20 years at UC Irvine, 7 years at UC Riverside, and a plethora of other prestigious positions in between. “He has a tremendous amount of administrative experience having served as an NSF Division Director, the Vice President of Research for Rutgers, and the Vice Chancellor of Research at UC Riverside, which makes him a great resource on how to navigate the challenges every organization faces,” adds Knoblock, Keston Executive Director of the Information Sciences Institute and Research Professor of Computer Science.
Pazzani’s career was sometimes shaped by unexpected opportunities. He recalls that “when Rutgers called and asked if I wanted to apply to be Vice President of Research, I was not sure I even wanted the job. But I decided to treat the interview as a practice interview. Turns out, they hired me!” Which should come as no surprise considering his achievements.
His early work was focused on personalization. In 2000, he formed a company to personalize news on mobile devices, where every millimeter counts on small screens. His customers were The Los Angeles Times and The San Diego Union Tribune. “We were very early in the space,” recalls Pazzani, whose student even built “the first personalization engine for Tivo, who was one of the first to personalize media. Now, Netflix and everyone else is doing it.”
Classic paper and salary predictions
One of his articles also won the AAAI Classic Paper Award in 2014 for a paper published in 1996, an honor given to pieces that are still a reference in the field decades after their publication. His most cited paper is about “trying to explain why something that should not work in practice actually does. This is the most mathematical work I have done.” When asked how he would explain it to a ten year old, Pazzani laughed: “You don’t.” His early work in explainable AI is more concrete: “We analyzed bias in machine learning systems so they produce things that people find insightful. For example, you can write equations to predict a baseball player’s salary from their statistics.”
Going forward, Pazzani is counting on ISI’s expertise in knowledge graph and computer vision to deepen his research. Skin cancer detection is on his radar, using a technology he developed to recognize bird species from a picture. “Today there are apps where you can take a photo of your mole, it analyzes it and tells you there is a 67% chance that it is cancerous. What I want to do is give you an explanation instead of a number: the borders are irregular and the color is this way, therefore it could be skin cancer and you should see a dermatologist.”
Yolanda Gil, research professor and director for major strategic AI and data science initiatives at ISI, first met Pazzani in the late 80s when she was still a student. She remembers him having “the foresight to work on topics that are still very important today for AI, such as explanation and causality” and expects him to help ISI reach a new level. “He will strengthen machine learning research at ISI and help us move USC forward more strategically in AI.”
Published on February 22nd, 2022
Last updated on March 10th, 2022