View article

Training discriminative classifiers involves learning a conditional distribution p(y ⁱ |x _i ), given a set of feature vectors x _i and the corresponding labels y _i , i=1...N. For a classifier to be generalizable and not overfit to training data, the resulting conditional distribution p(y _i |x _i ) is desired to be smoothly varying over the inputs x _i . Adversarial training procedures enforce this smoothness using manifold regularization techniques. Manifold regularization makes the model's output distribution more robust to local perturbation added to a datapoint x _i . In this paper, we experiment with the application of adversarial training procedures to increase the accuracy of a deep neural network based emotion recognition system using speech cues. Specifically, we investigate two training procedures: (i) adversarial training where we determine the adversarial direction based on the given labels for the training data and, (ii) virtual …