Computer Vision News - July 2020

Trevor Darrell 44 was so exciting. What’s fascinating is that it’s still a very active challenge. We haven’t solved it, although we’ve made significant dents in it as a field. In fact, it used to be an off-on-the-side topic in mainstream computer vision, but now everybody is acknowledging and discussing the fact that supervised learning under IID conditions is just such a narrow problem statement that it’s not what’s exciting today. Another example would be work with a former student, Lisa Anne Hendricks , on explainable AI. There, the realization that we came to was not to think of the goal being to make somebody feel good or trust the system, but that adding explanations to a system can make the system perform much more accurately. The way we think of it now is that explanations are almost another modality or stream of observations or supervision. It’s a little over-simplified perhaps, but I do think that the way humans learn is by being taught, and we teach humans not just with examples, but with stories. When we test humans, we almost always ask them to show their work. Machine learning regimes which strongly encourage or require the student to show their work seem to be more effective. That’s another theme that we’ve been exploring in our recent work at Berkeley. While we’re on the subject of former students and postdocs, I would also like to say that we’re super excited to have While we’re on the subject of former students and postdocs, I would also like to say that we’re super excited to have Angjoo Kanazawa joining us as a new faculty member at UC Berkeley! We’re really looking forward to seeing the fantastic group that she’s going to grow with us. Now that we have talked about what you have learned from your students, how about what you have learned from your own professors and teachers over the years? Is there anything that they have taught you that you’d like to tell our readers about? The earliest mentor I had was a remarkable woman, Ruzena Bajcsy. What an incredible impact she has made on the field. She was a professor at the University of Pennsylvania’s GRASP lab “Everything I’ve learned is generally from one of my students.” Former student Lisa Anne Hendricks Best of CVPR 2020