Fig. 3
From: Unsupervised deep learning reveals prognostically relevant subtypes of glioblastoma
Model selection cross-validation reconstruction error. For all plots in this figure, the y-axes represent reconstruction error. (a) Reconstruction error for multiple sizes of a one hidden layer network. The size of the 1st hidden layer increases by 200Â units for each consecutive curve. (b) Reconstruction error for two hidden layer networks. (c) Reconstruction error for three hidden layer networks. The blue rectangle represents network architectures that overfit the data. (d) Reconstruction error for four hidden layer networks