This project is funded by WASP (Wallenberg AI Autonomous Systems and Software Program). We want to explore if the advantages of equivariant neural networks over regular convolutional neural networks hold for a larger set of experimental conditions. To perform this, we want to vary the hyperparameters of the optimization process, the underlying architectures, the datasets, and the symmetry groups for the equivariant operations. Previously, a dataset for oral cancer classification and another for semantic instance segmentation of cells in various conditions have been looked at. To begin with, we would like to repeat the experimental procedure on another dataset, consisting of electron microscopy pictures of different species of viruses. Investigating different symmetry groups is the second priority. Finally we would like to investigate different network architectures and optimization hyperparameters.