New technologies have enabled the development of granular calorimeters with millions of channels. The signal of the ultra-sampled shower produced by these devices is thought to provide greater discriminating power to event reconstruction. Combining sub-nanosecond digitization with small area photosensors in a fiber calorimeter, we propose an enhancement to the traditional dual readout design that provides benefits of both high-granularity and multi-readout. We show that by applying machine learning techniques, namely Convolutional Neural Networks, Graph Neural Networks, and Recurrent Neural Networks to both a highly granular and proposed fiber calorimeters. We see, for instance, in the simple high granularity setup, the CNN improves reconstructed energy resolution from ~40 to ~33 %/sqrt(E). These results indicate the spatial distribution of energy deposition within the sensitive elements is both identifiable (able to be learned) and representative of underlying physical processes.
|TIPP2020 abstract resubmission?||No, this is an entirely new submission.|