Artificial Intelligence

This course presents the fundamentals of statistical learning theory, from fundamental limits, to algorithms, their analysis and practical implementation. Important concepts such as Vapnik-Chervonenkis theory, Empirical Risk Minimization, Stochastic Gradient Descent algorithms and Kernel methods will be presented in details. A lab session will enable students to apply the theoretical contents to data.

This course will provide students with the principles of representation learning and deep learning by covering the following subjects: Neural Networks, Backpropagation and stochastic gradient optimisation, Auto-encoders, Hyper-parameters and training tricks for neural networks, regularization, Deep Belief Networks and Deep Boltzmann Machines. Students will apply these approaches during a practical lab session.

Networks (or graphs) have become ubiquitous as data from diverse disciplines can naturally be mapped to graph structures. Characteristic examples include social networks (e.g., Facebook, Twitter), information networks (e.g., the Web) as well as technological networks (e.g., the Internet). The problem of extracting meaningful information from large scale graph data in an efficient and effective way has become crucial and challenging with several important applications and towards this end, graph mining and learning methods constitute prominent tools. The goal of this course is to present recent and state-of-the-art methods and algorithms for analyzing, mining and learning large-scale network data, as well as their practical applications in various domains (e.g., the web, social networks, recommender systems).

In this course students will survey the major aspects of Artificial Intelligence usage and techniques in the context of medical imaging. Medical imaging is a fascinating and major application area for Artificial Intelligence/Machine Learning where the stakes are high and the impact on society can be lasting and strong. However, this is also an application area fraught with challenges. In particular, is it important that AI results in this area be trustworthy, repeatable and auditable. Physicians are usually not happy with black-box decisions that come with no explanations. Fortunately many novel techniques exist to produce exactly this kind of results, which will be discussed in detail.

The high number of documents as well as bodies created to deal with “the ethics of AI” leads us to wonder why AI has recently become a particular object of attention and which ethics is at stake. Starting from a review of international documents and recommendations, we will highlight issues with the vocabulary used, misleading postulates so as tensions and paradoxes. As an example, we will particularly focus on the “human control” principle.  We will then insist on the risks of misuse of ethics and the need for a true ethical reflection going with AI research so as with design and use of AI systems.

According to the World Health Organization, cardiovascular diseases (CVDs) are in 2021 the number one cause of death globally, taking an estimated 17.9 million lives each year. Access to essential noncommunicable disease medicines and basic health technologies in all primary health care facilities is essential to ensure that those in need receive treatment and counselling. However, more than 75% of CVD deaths occur in low and middle income countries, where this access is not guaranteed, and trained clinicians are only available in small numbers. Portable technologies such as electronic stethoscopes and point-of-care ultrasound probes, when enhanced with artificial intelligence and computer assisted decision systems, could be the key to a more successful management of cardiac disease in underprivileged environments.