Computer Vision News - February 2018

Computer Vision News 29 Michela Paganini to decide what to work on because people always come to me with a lot of brilliant ideas and a lot of cool projects that one could get involved with. Unfortunately, there is not enough time to think carefully about all of them. Now we are curious, too! What did you learn about the universe that we don’t know? What matter is really made of? Well, to be fair, before I even joined CERN, certainly the biggest discovery in recent years has been the Higgs boson. That was in 2012, and I was still an undergraduate student, not working in physics. I can’t claim that as one of my victories! We’ve learned a lot more about how to distinguish certain signatures of particles in our detector. What we’re trying to do as experimentalists is to try and go look for those signatures. The issue that we have to deal with is that we cannot observe those particles as they come to be at the center of our collision point; we have to measure their decay products through the detector. As you can imagine, a detector is something with a certain finite granularity. A lot of the information gets lost in a way as particles propagate through the detector. The real questions that we are trying to answer are how do we best extract information from what we can actually measure, which is finite in a way. The machine learning is helping us answer how we can detect particles that we know very, very well such as bottom quarks or charm quarks. How do we distinguish the two once they interact with the detector? They look very similar! They have similar properties, even though they are different particles. How does your work involve Computer Vision? In ATLAS we use computer vision techniques to identify particles instead of cars, objects, or pedestrians. Imagine our detector as a large high-definitional camera that takes pictures of the collisions: our data can then be transformed into image format and analyzed using CV-inspired methods. This is a relatively new way of looking at LHC data which brings us closer to detector-level information; previously, instead, this information was simply summarized into a set of engineered features. In recent years we have seen an exponential growth in papers in particle physics using convolutional neural networks on image representations of our data, but the question of how to best represent the unique nature of our datasets (whether as images, sequences, graphs, etc.) is still very much an open research question, with lots of recent contributions presenting new and innovative strategies. You’re Italian. Yes, I am Italian. I grew up in a city called Busto Arsizio, near Milan. Women in Science In the ATLAS cavern, in front of one of the endcap wheels of the ATLAS detector. Michela is proudly wearing the jersey of Pro Patria, the football team of Busto Arsizio.