Fig. 2From: AutoDTI++: deep unsupervised learning for DTI prediction by autoencodersFeed-Forward/Backward process is shown for denoising autoencoder. The input is obtained from the matrix of interactions, unknown values are turned to zero, some interactions input are corrupted, and a dense estimate is finally constructed. Before back-propagation, unknown interactions are converted to zero error. Use \({\varvec{\beta}},\boldsymbol{ }\boldsymbol{\alpha }\) hyper-parameters, reconstruction, and prediction errors are reweighedBack to article page