Deep Learning for Signal and Image Processing

The course will present recent artificial intelligence and machine learning techniques for multi-dimensional signal processing and pattern recognition, with a specific focus on Deep Learning (DL) approaches. With respect to traditional pattern recognition algorithms, DL methods have the advantage of automatically extracting distinctive data representations from multidimensional signals, thus reducing the need of domain expertise in a specific field. Currently, DL approaches represent the state of the art in several fields, such as industrial monitoring, medical imaging, biometric recognition, object classification, and ambient intelligence. However, the choice of the best DL model for a specific application is still a challenging design aspect. The course will present an overview on the main DL approaches for signal and image processing, such as Convolutional Neural Networks, Autoencoders, and Generative Adversarial Networks. Then, the course will present application examples for heterogenous scenarios, including industrial monitoring and ambient intelligence.