Workflow from Scientific Research

A graph-convolutional autoencoder is trained and applied to our set of CMR-derived LV meshes (number of vertices M = 5,220) to produce low-dimensional representations of these shapes. In each layer, a representation with fewer vertices is obtained. The bottleneck z r of the autoencoder with hyperparameters r is a {n}_{z}^{r} -dimensional vector for each run r ( {n}_{z}^{r in {8,16} ). ReLU, rectified linear unit.

DOI