Speaker: Sergio Álvarez
Abstract: Deep Neural Networks (DNNs) have revolutionized many fields in pattern recognition like speech recognition and object detection. There are, however, some applications in which Neural Networks struggle to offer competitive performance, mainly sensitive ones. These applications require to account for uncertainty in the predictions, e.g. the true probability of a patient not having cancer. One way of modeling uncertainty is to take a Bayesian approach, yet this is especially challenging in the case of DNNs. In this presentation we review some common approaches in the literature to approximate Bayesian Neural Networks (BNNs) and elaborate on how we introduce uncertainty in predictions.