Using Unsupervised Machine Learning to make new discoveries in space data

Modern research in space science is accumulating an exponentially growing amount of data. The size of the datasets produced by simulations has grown in time with the speed of the computers, following its own Moore’s law. At the same time, also data from space mission has grown, with more images, more time series and more complex datasets such as spectrograms and velocity distribution function. Mining information and making discoveries out of this wealth of data has far transcended the ability of the human mind. Artificial intelligence (AI), instead, has taken off in many areas of research and applications. Space science can benefit from this new trend. We report here on the recent developments on the application of AI to the analysis of space data. Without limiting ourselves to it, we focus especially on the work done in the context of the AIDA Horizon 2020 project (Artificial Intelligence for Data Analysis, www.aida-space.eu). We show how AI developments are transforming the way we analyze space data. In particular, we find promise in unsupervised Machine Learning (ML) to give optimism for fundamental new discoveries. Supervised ML can expand the range of known methods of analysis, learning from human training and treat vast amount of data. But unsupervised ML, a method where the learning is not guided by previous knowledge, can truly find unexpected new discoveries. Unsupervised ML provides techniques to treat large data sets and discover within them features, correlations and physical laws that would escape traditional approaches. In this chapter we review some of the applications of unsupervised ML developed by AIDA.