In recent years, artificial neural networks have become a standard tool in many analyses to increase the sensitivity of measurements and largely replaced other multivariate techniques. The hyperparameters of the neural network, e. g. the number of hidden layers in a multilayer perceptron, are however usually chosen based on intuition and experience without any optimization. Additionally, the absence of overtraining is often only verified by visually inspecting the network’s output distributions.

In this talk, a framework to perform automated hyperparameter optimization with a special focus on directly including objective overtraining conditions as part of the optimization is presented. Furthermore, its first application in the ATLAS vector boson polarization analysis of W^±W^± scattering is discussed.