Computer Vision News - April 2022

22 Women in Science theory to address this. I’ve been interacting with people from YouTube and with people from Google Books via DeepMind applied research: these people have concrete product constraints. They come to me and say, “ We have this problem. We want this, and we want that. We have this constraint and that constraint .” I’m like, “ Okay, we need to sit down andwrite downwhat these things mean in math. Then we can write down the equation and find an algorithm that is going to work. ” I really enjoy this modeling part. This is what I have been working on since the beginning of my PhD. I was working with Criteo and with Peugeot in France to help them formulate their problem. Then I came up with a theoretical problem, which is kind of a toy problem but models the situation. Then the eureka moment comes at the end when we try to prove that our algorithm actually makes sense and works. Passing this knowledge into application takes much more time. Do you ever get out of the office and see something that does not work as it should and think, “ How do I translate this into a not always the same. Sometimes there is traffic. Sometimes there is a trash truck in one street, so you want to go to another street. The best option is changing with time. The idea is to detect that and adapt. There are real applications to non- stationary time series or non-stationary environments. Personally, I don’t directly work on one application in particular. I’m giving practitioners tools to think about if they should address their problem this way, or that way. It depends on the conditions that I can formulate in my papers. But I don’t go and solve one particular problem. Does this mean that you enjoy solving a problem more than seeing your work translated into real-world applications? There are two things. There is the eureka moment, but I would say that this is the last part. The first part that I enjoy a lot is looking at real-world problems and asking the question: How can I write it in math? How can I translate it into a model that I can solve? I know that we have the tools in statistics, in mathematics, and in learning I want to track the non-stationary, this kind of volatility of the environment.