While lens search with convolutional neural networks became the state of the art technique, the work carried out by the Marie Curie Fellow goes beyond this in multiple aspects. First, normally a stringent pre-selection is done before applying the convolutional neural network. The published work is the first that analyses more than 100 million images without any stringent cut beforehand. This is crucial for the ongoing and upcoming wide filed surveys, such as the Euclid space mission of Europe, delivering billions of images that need to be analyzed. Second, the work includes for the first time an environment analysis. Tested against visual inspection and known non-lensing galaxy clusters, a new method was developed using photometrically determined redshifts. These redshifts were obtained by combining results from three different techniques to increase their precision, of which one algorithm, based on deep learning, was previously developed by the Fellow himself. Third, the publication includes a first total mass model for all identified lens candidates, obtained in a fully automated way, making it to the larges sample ever modeled in a uniform way. Although these rough mass model is not enough for a detailed analysis and comparable with the model required for the lensed supernovae, automated and fast detection, environment analysis, and modeling will be crucial in the upcoming decade as the European space telescope is expected to deliver on the order of 100 000 galaxy scale lenses.
On the other hand, improving the total cluster mass model for their details analysis is the other task tackled by the Marie Curie Fellow. Including the observed pixel values of the lensed supernova host and reconstructing its unlensed light distribution increases the constraints by around three orders of magnitude, leading to a new generation of mass models with unprecedented local accuracy. However, this increase of data results in significant computational challenges.
