Manifold learning for image-based digital twining in mechanics of materials

This is a Data Challenge.

The aim of the Data Challenge is to predict an image of mechanical fields around defects from the image of these defects. It is a regression task by using supervised machine learning. Each prediction must also generate a coordinate of the input defect in a latent space of reduced dimension. Such a latent space can be further used for model order reduction in mechanics of materials, data interpolation, clustering… But the exploitation of the latent space is not in the scope the Data Challenge. Here, the smaller the dimension of the latent space for a given prediction accuracy, the better.