Recent rapid developments in Deep Learning, which can implicitly capture structures in high-dimensional data, will lead to the opening of a new chapter of astronomical data analysis. As a new implementation of Deep Learning techniques in the fields of astronomy, we here report our application of Variational Auto-Encoder (VAE) using deep neural network to spatio-spectral analysis of the data from the Chandra X-ray Observatory, in the particular case of Tycho’s supernova remnant. Previous applications of Machine Learning techniques to the analysis of SNRs have been limited to principal component analysis (Warren et al. 2005; Sato & Hughes 2017) and clustering without dimensional reduction (Burkey et al. 2013). We have implemented an unsupervised learning method combining VAE and Gaussian Mixture Model (GMM), where the reduction of dimensions of the observed data is performed by VAE and clustering in the feature space is by GMM. We have found that some characteristic features such as the iron knots in the southeastern region can be automatically recognized through this method. Our implementation exploits a new potential of Deep Learning in astronomical research.
Link to PDF (may not be available yet): P1-3.pdf