“…56 The MultiTaskClassifier model is a fully connected deep residual network for multitask classification composed of preactivation residual blocks. Hyperparameterization for this first model was performed to optimize the best parameters, including layer size [256,256], [256,256,256], [256,256,256,256], dropout [0.0, 0.1], learning rate [0.005, 0.0035, 0.0025, 0.0010], number of epochs [5,10,20,30,40,50], momentum [0.0, 0.9], decay [0.0, 0.1], and batch size [32,64,128]. This was done using the GridHyperparamOpt optimizer, 67 the hyperparameter search method, 61 and the ROC-AUC metric.…”