Deep convolutional neural networks with an application to visual recognition

This thesis provides an overview of Multilayer perceptron model by explaining, as far as possible, how it works, what are its properties, its limits and possible tricks to improve it. Furthermore will be explored the new learning paradigm known as Deep Learning. Finally another deep neural network architecture called convolutional neural network is introduced, since, today, this kind of newtork is the state-of-the-art for visual recognition. The goal of this thesis, indeed, is to lay down the foundations of a tool for localization within indoor environments, where location estimation is not possible by using GPS signal. Deep convolutional neural networks turns out to be the better choice in order to develop this task.