IRIS-HEP Fellow: Raghav Kansal
Fellowship dates: Jun-Aug 2019
Home Institution: University of California, San Diego
High granular calorimeters will be the biggest novelty of the CMS Phase II upgrade and, in general, for the next generation of collider experiments. This kind of detectors offer more opportunities but much more complexity for ordinary tasks such as detector simulation. In order to stay within the technical budgets (e.g. computing time) and satisfy the demand for large simulation samples, experiments will have to work on faster and more accurate simulation techniques. Deep Learning, and in particular generative models, offer an interesting possibility to speed up the simulation technique. Moreover, Deep Learning solutions are particularly suitable for HGCAL, given the pixelated nature of the problem. This project aims to adapt existing work about GAN for fast simulation to the irregular geometry of this detector, using graph networks as a way to learn a sparse representation of the hit distribution and embed it in a regular array, where traditional computing vision techniques can be used.