Fig. 1

Hyperparameter optimisation for autoencoders using Pareto analysis. Left panel: PCA visualisation of the four evaluation metrics (i.e. ARI, NMI, FM and Jaccard) on each of the four optimisation datasets. Each point corresponds to a single combination of hyperparameter values including random projection size, encoded feature space size, and autoencoder learning rate during backpropagation; each combination/point is colour-coded by the number of times it was assigned Pareto rank 1 (i.e. the combination that gives best clustering performance) across all possible combinations of the four optimisation datasets. Right panel: Autoencoder architecture as determined by the hyperparameter optimisation procedure