In a large feedforward neural network, overfitting can be greatly reduced by randomly omitting half of the hidden units on each training case. This prevents complex co-adaptations in which a feature detector is only helpful in the context of several other specific feature detectors. Instead, each neuron learns to detect a feature that is generally helpful for producing the correct answer given the combinatorially large variety of internal contexts in which it must operate. Random “dropout” gives big improvements on many benchmark tasks and sets new records for object recognition and molecular activity prediction. The Merck Molecular Activity Challenge was a contest hosted by Kaggle and sponsored by the pharmaceutical company Merck. The goal of the contest was to predict whether molecules were highly active towards a given target molecule. The competition data included a large number of numerical descriptors generated from the chemical structures of the input molecules and activity data for fifteen different biologically relevant targets. An accurate model has numerous applications in the drug discovery process. George will discuss his team's first place solution based on neural networks trained with dropout.