Cosmic super string detection using dilated convolutional neural network with focal loss

Abstract

Cosmic string are objects of great importance and investigation for cosmic string has been done from last 20 years. There are a lot of models to detect cosmic string.But a very few are to detect the location of cosmic string.We propose a framework to detect the location of cosmic string. We used di- lated convolutional net with focal loss instead of cross-entropy to improved the performance of the framework on weak samples. The neural network we trained is able to detect and locate cosmic string on noiseless CMB temper- ature map down to a string tension of less then G =5 10􀀀9. We expect to use more accurate simulation to produce data set to improve the con dence of the model.