Scalable deep convolutional neural networks for sparse, locally dense liquid argon time projection chamber data

Deep convolutional neural networks (CNNs) show strong promise for analyzing scientific data in many domains including particle imaging detectors such as a liquid argon time projection chamber (LArTPC). Yet the high sparsity of LArTPC data challenges traditional CNNs which were designed for dense data such as photographs. A naive application of CNNs on LArTPC data results in inefficient computations and a poor scalability to large LArTPC detectors such as the Short Baseline Neutrino Program and Deep Underground Neutrino Experiment. Recently, submanifold sparse convolutional networks (SSCNs) have been proposed to address this class of challenges. We report their performance on a three-dimensional (3D) semantic segmentation task on simulated LArTPC samples. In comparison with standard CNNs, we observe that the computation memory and wall-time cost for inference are reduced by a factor of 364 and 33, respectively, without loss of accuracy. The same factors for 2D samples are found to be 93 and 3.1, respectively. Using SSCN and public 3D LArTPC samples, we present the first machine learning-based approach to the reconstruction of Michel electrons, a standard candle for energy calibration in LArTPC due to their very well-understood energy spectrum. We find a Michel electrons identification efficiency of 93.9% and a 96.7% purity. Reconstructed Michel electron clusters yield 95.4% in average pixel clustering efficiency and 95.5% in purity. The results are compelling in showing the strong promise of scalable data reconstruction technique using deep neural networks for large scale LArTPC detectors.