Finalità

The course aims to illustrate the fundamental concepts underlying the most recent Artificial Intelligence techniques for  multimedia. In particular, deep neural networks will be presented based on current Deep Learning approaches, paying particular attention to generative models.

The goal is to explore the dimension of creativity by tacking tasks that were impossible only until recently: the creation of artificial images or human faces, the application of a pictorial style on a  sample image (neural style transfer), the creation of a Question&Answer generator, composing text paragraphs or music scores.

Experiments on generative models will be performed in class through the use of the Python programming language and Deep Learning libraries such as Keras and Tensorflow.

Part I: Fundamentals

– Neural networks

– Introduction to Deep Learning

– Convolutional Networks

– Introduction  to Generative Models

– Variational Autoencoders

– Generative Adversarial Networks

Part II: Applications

– Generation of artificial images

– Neural Style Transfer, DeepFake

– Text generation

– Composing music

 

Prerequisites

Requirements: Basic programming skills required. Knowledge of an object-oriented programming language is a plus.

Exam modality

The exam will be based on a project and oral discussion.

Books:

Generative Deep Learning: Teaching Machines to Paint, Write, Compose,
and Play, O’Reilly, 2019.

Rashid, Tariq. Make your own neural network: a gentle journey through
the mathematics of neural networks, and making your own using the
Python computer language. CreateSpace Independent Publ., 2016.

Goodfellow, Ian, et al. Deep learning. Vol. 1. Cambridge: MIT press, 2016.

Deep Learning with Keras: Implementing deep learning models and
neural networks with the power of Python, Packt, 2017.