9 Gitta Kutyniok Computer Vision News Gitta, but you also have many, many years in front of you of teaching and researching. What would be an ideal goal to attain for you? One direction where I’m working seriously is in making deep neural networks and AI reliable. It turned out, and this relates to what I said about limitations, that one main limitation is the hardware we train it on. We use digital hardware like CPUs and GPUs, so you can show that there are actually problems concerning the computability of these types of algorithms. You can model digital hardware as a Turing machine, and you can get rigorous results. What you also see is that, for instance, in self-driving cars, we are not as far as we thought. These are also indications of this problem. One future direction we also work in is to augment digital hardware by analog hardware. Analog hardware like neuromorphic computing, quantum computing, biocomputing, and developing algorithms for those, and then showing that those are reliable. In that sense, that’s one key goal that many people in my research group work on. If you ask me what I hope to achieve within the next 5-10 years, this is something I would like to achieve. For different application areas like medicine, robotics, and telecommunication to build these augmented hardware platforms and algorithms and show that this is reliable and satisfies the constraints given by the EU Act and by the G7. This is something which is, I’d say, a vision for the future. What has been the biggest eureka moment in your career so far? Wow, that’s actually very, very, very difficult. For me, when I entered the area of explainability, at first, it seemed like there was no mathematics possible in some sense. It’s an area which has a lot of social and psychological components. It’s not clear what an explanation is. But now, also doing the CVPR paper, we have some first mathematical results. That was something which really surprised me and, in that sense, was maybe something like such a moment. What is the latest CVPR paper? The latest paper was about a novel explainability approach using ideas, many from mathematics like applied harmonic analysis, using wavelets and shearlets, and getting Watch her interview in video!
RkJQdWJsaXNoZXIy NTc3NzU=