This paper illustiates how the statistical structure of natural signals may help understand cognitive phenomena. We focus on a regularity found in audio visual speech perception. Experiments by Massaro and colleagues consistently show that optic and acoustic speech signals have separable influences on perception. From a Bayesian point of view this regularity reflects a perceptual system that treats optic and acoustic speech as if they were conditionally independent signals. In this paper we perform a statistical analysis of a database of audiovisual speech to check whether optic and acoustic speech signals are indeed conditionally independent. If so, the regularities found by Massaro and colleagues could be seen as an optimal processing strategy of the perceptual system. We analyze a small database of audio visual speech using hidden Markov models, the most successful models in automatic speech recognition. The results suggest that acoustic and optic speech signals are indeed conditionally independent and that therefore, the separability found by Massaro and colleagues may be explained in terms of optimal perceptual processing: Independent processing of optic and acoustic speech results in no significant loss of information.

An active object recognition system has the advantage of acting in the environment to capture images that are more suited for training and lead to better performance at test time. In this paper, we utilize deep convolutional neural networks for active object recognition by simultaneously predicting the object label and the next action to be performed on the object with the aim of improving recognition performance. We treat active object recognition as a reinforcement learning problem and derive the cost function to train the network for joint prediction of the object label and the action. A generative model of object similarities based on the Dirichlet distribution is proposed and embedded in the network for encoding the state of the system. The training is carried out by simultaneously minimizing the label and action prediction errors using gradient descent. We empirically show that the proposed network is able to predict both the object label and the actions on GERMS, a dataset for active object recognition. We compare the test label prediction accuracy of the proposed model with Dirichlet and Naive Bayes state encoding. The results of experiments suggest that the proposed model equipped with Dirichlet state encoding is superior in performance, and selects images that lead to better training and higher accuracy of label prediction at test time.