Particle Flow (PFlow), the task of reconstructing final state particles from the detector readout, is at the heart of Collider physics. The task is quite challenging due to the complex nature of the detector data, which is heterogeneous (tracks, cells, cluster of energies), sparse, and usually originates from a non-uniform detector geometry. The recent advancements in the felid of Machine learning (ML) have suggested new possibilities beyond the scope of traditional parametric approaches to solving this complex task. In the talk, I will show models that use a set of detector hits as inputs and predict a set of reconstructed particles. I will also discuss a few novel approaches with graph-based architectures, and understand how we can combine ML with physical motivations to obtain State-of-the-art performance. I will also discuss how these models are well-suited to tackle generative problems.

Per connettersi a zoom
Topic: Theory and Pheno seminars
https://infn-it.zoom.us/j/8573185271
Meeting ID: 857 318 5271