演講資訊

摘要：

Astronomy today is in an exciting and challenging era with the fast-increasing amount of data from large-scale sky surveys and future facilities. This necessitates the development of machine learning techniques in astronomical studies to help improve conventional analyses in various aspects, as well as inspire new perspectives to look into data. In this talk, I will share my experience of machine learning techniques, in particular, on galaxy morphology. We construct one of the largest galaxy morphological classification catalogues including over 20 million galaxies for the Dark Energy Survey using a convolutional neural network. Galaxy morphology is further explored by unsupervised machine learning to avoid human’s opinion in categorisation. With our novel approach, galaxies in the local universe are categorised into 27 classes with distinctive physical properties. In addition to the improvement in efficiency and accuracy of astronomical analysis, I will also discuss some problems we faced in these works with machine learning approaches.