PyTorchDIA: a flexible, GPU-accelerated numerical approach to Difference Image Analysis

ABSTRACT We present a GPU-accelerated numerical approach for fast kernel and differential background solutions. The model image proposed in the Bramich Difference Image Analysis (DIA) algorithm is analogous to very simple convolutional neural network (CNN), with single filter (i.e. kernel) an added scalar bias background). Here, we do not solve discrete pixel array classical, analytical linear least-squares sense. Instead, by making use of PyTorch tensors (GPU compatible multidimensional matrices) associated deep learning tools, via inherently massively parallel optimization. By casting DIA problem as optimization that utilizes automatic differentiation our both flexible choice objective function, can perform on astronomical data sets at least order magnitude faster than its classical analogue. More generally, demonstrate tools developed machine be used address generic analysis modelling problems.