Latent Space Explorer: Unsupervised Data Pattern Discovery on the Cloud

Abstract

Extracting information from raw data is probably one of the central activities of experimental scientific enterprises. This work is about a pipeline in which a specific model is trained to provide a compact, essential representation of the training data, useful as a starting point for visualization and analyses aimed at detecting patterns, regularities among data. To enable researchers exploiting this approach, a cloud-based system is being developed and tested in the NEANIAS project as one of the ML-tools of a thematic service to be offered to the EOSC. Here, we describe the architecture of the system and introduce two example use cases in the astronomical context.