Cosmic-Ray Energy Reconstruction in IceCube Using a Convolutional Neural Network with Low-Level Inputs

The purpose of our research is to build a convolutional neural network capable of energy reconstructions using data detected at the IceCube Observatory for use with high-statistics, minimally-cut cosmic-ray anisotropy studies. Our goal is to use simulation from the IceTop surface array to build a lightweight model that can successfully reconstruct contained and uncontained events over a large zenith range. By using only low-level charge and time information as inputs, we hope to minimize the amount of systematic uncertainty in our model while maintaining the accuracy of models whose input includes higher-level parameters. Current research primarily focuses on improving our model by normalizing input and clipping time data, as well as introducing rotational invariance by rotating the simulated events. Our current model is capable of estimating 68% of unfiltered simulations within 15% of their true energies.