AIdra_abstract.txt

Predictions on the ML Cup blind test were obtained by ensembling (bagging) 32 models, trained on data sets obtained by random 
sampling with replacement on the entire training data set (final retraining after model assessment). All models used in this 
technique share the same hyperparameter configuration, NN's topology and Learning Algorithm, according to the result of two iterations 
of 3-fold CV (coarse-grained followed by fine-grained). The NNs considered were trained with Standard Backpropagation. 
The reference topology consists of a single hidden layer with 32 hidden units and a sigmoid activation function.