CVPR Daily - Monday

Judy Hoffman is an Assistant Professor in the School of Interactive Computing at Georgia Tech and a member of the Machine Learning Center. Read 100 more interviews with women in Computer Vision Judy, can you tell us about your work? My work is in computer vision and machine learning. I focus a lot on trying to make sure that visual recognition systems are robust and reliable. I focus explicitly on tasks like domain adaptation and adversarial robustness, trying to understand how systems might fail when assumptions change about the input data distribution. How did you get involved in this type of work? I got here in a little bit of a roundabout way. I first started doing research as an undergraduate in Ken Goldberg’s lab at the University of Berkeley. I was working on robotics and a little bit on medical imaging for surgical robotics. I found the imaging part extremely fascinating and decided that for my PhD, I wanted to focus more in that area. Early on in the program, I saw a talk by Kate Saenko, who was a postdoc at the time. She was talking about domain adaptation. I found that problem fascinating. It seemed really important to me and a nice, broad, open area. So I decided to work on that for my PhD. During the last few years, I’ve kind of expanded to think about this problem of robustness more broadly. Where adversarial robustness as well as data distribution might impact certain populations in disparate ways has also become interesting to me in the last few years. Before you decided to go in that direction, what were you planning to do? What were your objectives before? Before, I was just getting exposed to different things. I was pretty open- minded about what I would work on, about where I would do research as a career and go to graduate school. I was just trying to grasp information at the time. My undergraduate research from the application focus was really motivating to work on things in the medical space. There was a great collaboration at the time with UCSF, where I actually got to go and visit clinical labs and even meet with some patients. 20 DAILY CVPR Monday Women in Computer Vision