New machine learning techniques for the analysis of large and deep, multi-color, multi-instrument, astronomical surveys – AstroDeep

This research project aims to develop advanced techniques for analyzing vast datasets from upcoming astronomical surveys, such as Euclid and LSST. These surveys aim to understand “Dark Energy”, the force behind the Universe’s accelerated expansion. They present specific challenges like handling overlapping objects in deep images, to measure weak gravitational shear and photometric redshifts. The team will leverage machine learning techniques, especially focusing on innovative neural network architectures, probabilistic deep learning and probabilistic analysis pipelines. Those techniques will efficiently process multi-petabyte datasets, with combined multi-instrument, multi color, pixel-level analyses.