Sorry, you need to enable JavaScript to visit this website.

This presentation introduces a Deep Learning model that performs classification of the Audio Scene in the subway environment. The final goal is to detect Screams and Shouts for surveillance purposes. The model is a combination of Deep Belief Network and Deep Neural Network, (generatively pre-trained within the DBN framework and fine-tuned discriminatively within the DNN framework), and is trained on a novel database of pseudo-real signals collected in the Paris metro.

In this paper, we propose a supervised subspace learning method that exploits the rich representation power of deep feedforward networks. In order to derive a fast, yet efficient, learning scheme we employ deep randomized neural networks that have been recently shown to provide good compromise between training speed and performance.

Deep convolutional network has been widely used in face recognition while not often used in face alignment. One of the most important reasons of this is the lack of training images annotated with landmarks due to fussy and time-consuming annotation work. To overcome this problem, we propose a novel data augmentation strategy. And we design an innovative training algorithm with adaptive learning rate for two iterative procedures, which helps the network to search an optimal solution.

We present a geometry-inspired characterization of
target response for active sonar that exploits similarity between
intra-class features to distinguish between different targets
against environmental objects such as a rock. Key innovation is to
represent feature manifolds as a set of ellipsoids, each of which
geometrically encompasses a unique physical characteristic of
the target’s response. We have demonstrated over experimental
field data that for a given target class, these feature ellipsoids