Courses

The ADSAI PhD program offers different types of training courses, some of them offered in collaboration with the Theoretical and Scientific Data Science (TSDS) PhD program at SISSA, as well as with the Master in Data Science and Scientific Computing (DSSC) at the University of Trieste.

Following regulations regarding the university educational credits (CFU) system, every year ADSAI students need to define a training program in collaboration wiht their supervisors; the program can involve ADSAI courses, partner courses and attendance to conferences/ schools.

ADSAI courses (2023-24)

ADSAI courses (advanced training on Artificial Intelligence and related disciplines)

SISSA TSDS courses (advanced training on Artificial Intelligence and related disciplines)

Please refer to SISSA TSDS PhD program webpages.

DSAI and SDIC courses (introductory training on Artificial Intelligence and related disciplines)

Please refer to UNITS DSSC MSc program, UNITS DSAI MSc program and UNITS SDIC MSc program webpages.

Other courses (general training)

Syllabi

Review of statistical modeling and inference (Leonardo Egidi, Francesco Pauli and Nicola Torelli)

The course is aimed at reviewing the fundamentals of statistical theory within the classical statistical paradigm. We review the whole setting of likelihood theory, with a special focus on maximum likelihood estimation, properties of maximum likelihood estimators and a brief review of hypothesis testing. We deepen then linear and generalized linear models, covering the basic theory of these modeling techniques, the main inferential tools and proposing applications with real and complex data.

Advanced topics in statistical modeling (Leonardo Egidi, Francesco Pauli and Nicola Torelli)

The course extends basic statistical models through the introduction of multilevel/ hierarchical models, semi and non-parametric regression models, spline functions, penalized likelihood approaches (such as LASSO and Ridge regression models), and hierarchical models for estimating smooth regression functions. Applications to complex datasets and simulation approaches will be illustrated during the course through the use of the Stan ecosystem and the statistical software R.

Population-based Optimisation Methods (Luca Manzoni)

The course “Population-based Optimization Methods” focuses on a collection of methods, generally inspired by natural phenomena, used for optimization and where multiple solutions to the same problems are iteratively improved. In this course we will explore genetic algorithms, genetic programming, evolution strategies, particle swarm optimization, differential evolution, how to represent particular problem domains (e.g., graphs and permutations), neuroevolution, common parallelization techniques, and a brief introduction to the runtime analysis of bio-inspired algorithms.

Biodiversity Informatics ( Stefano Martellos)

The course aims at providing basic knowledge to organize, manage, and use biodiversity databases, both at management and research level. Furthermore, it aims at providing knowledge to support students in the drafting of research projects and proposals. Specifically, the course will provide: knowledge of the different types of data that exist in the field of biodiversity; understanding of the modern approached form digitizing, aggregating, and managing different types of biodiversity data; understanding of species distribution models; knowledge of the main tools for the production of interactive digital identification keys, and understanding of their logic. At the end of the course, students will be able to organize biodiversity data in databases, and aggregate them in federated and / or interoperable systems. They will also be able to use data in different lines of research, from ecological niche modelling to taxonomy, and therefore will be able to interact with different research groups in multidisciplinary approaches. Learning skills are stimulated by the use of several scientific papers and other materials, thus leading to a more critical and in-depth learning.

Advanced topics in Reinforcement Learning (Antonio Celani)

This short course will address issues in Multi-agent reinforcement learning, from Markov games and the connection with game theory to multi-armed bandit algorithms.

Advanced topics in network analysis: theory and applications (Domenico De Stefano)

Network analysis is a set of statistical techniques focused on uncovering and modeling patterns of relationships that arise from the interactions of people, groups, organizations, or any other entities. Network analysis is an in-demand approach to data analysis because it identifies underlying structures within data characterized by complex dependence structures. This short course will examine the most recent developments in statistical methods for analyzing and modeling the structure and dynamics of complex networks. The focus will be on methods for identifying patterns of relationships within large-scale datasets, predicting the probabilities of ties between nodes, modeling the dynamics of evolving networks. Applications will be drawn from computational social sciences and other disciplines.

Evolutionary Robotics (Eric Medvet)

The course introduces the main concepts of evolutionary robotics, that is, the family of optimization techniques based on the paradigm of natural evolution (evolutionary computation) and the principle of embodied cognition. Moreover, it surveys recent and significant research works on evolutionary robotics, highlighting their methodological aspects and their potential applicability to other domains. Concerning evolutionary computation, beyond key components, the course covers its application to the optimization of neural networks. Finally, the course introduces the reality gap problem and briefly surveys the main mitigation strategies.

Optimization-based control (Felice Andrea Pellegrino)

The course on ‘Optimization-based control’ will deal with optimization applied to control of linear, possibly constrained, discrete-time dynamical systems. After an introduction to the linear discrete-time dynamical systems and a survey of the relevant basic results of systems theory, the focus will be on (i) How to employ optimization to get open-loop optimal control sequences,(ii) How to get closed-loop control laws from open-loop control sequences, (iii) How to employ online optimization to control dynamical systems; (iv) How to employ optimization to get open-loop optimal control sequences in a model-free, data-driven fashion.

From Data to Models: An Introduction to Health Research (Andrea Rocca, Giulia Barbati - University of Trieste - and Daniela Zugna - University of Torino)

In the first part of the course the variety of data in health research will be presented, from structured “tabular” data to complex omics data, including also unstructured forms such as images, signals, and text. Emphasis is placed on the critical need to understand the data generation process and its significance in addressing biologicalchallenges. Then, the diagnostic medical process will be presented, highlighting the way medical doctors formulate diagnosis and evaluate treatment options. In the second part, the course will provide a concise overview of key study designs in epidemiology and clinical research, including Randomized Controlled Trials (RCTs) and Observational Studies. The choice of the design is closely linked to the specific scientific question to be addressed and the practical feasibility. The statistical tools used in prognostic modeling will also be briefly presented. The focus will be on the essential steps involved in constructing a prediction model, from the selection of predictors to the processes of internal and external validation.In the third part, an introduction to causal inference approaches in health research will be presented, focusing on defining the causal relationships and confounding roles of the factors under study with respect to an outcome of interest, and the key difference between the “association” and “causation” concepts will be discussed

Copula modeling – Extreme value theory (Francesco Pauli and Roberta Pappadà)

The course will give a brief introduction to univariate methods for extreme value modeling (generalized Pareto distribution, Poisson point process) with and without covariates. The course will offer a basic introduction to copulas for dependence modelling and their main properties, along with the most important theoretical results. Applications to relevant fields will be illustrated: computation of Value at Risk in finance; Risk Aggregation; Financial time series modelling with copulas.

Artificial Intelligence in society (Simone Arnaldi)

Artificial intelligence (AI) holds the promise of advancing society in a number of ways. At the same time, AI is increasingly implicated in controversies around its applications. Such an ambivalent public image of AI has led scientific societies, firms, governments and regulators to become more active in issuing blueprints, strategies, and regulations aimed at steering AI towards desirable social goals, while averting negative impacts on society. The Module will discuss the societal implications of AI, by reviewing the social and ethical aspects of this field and by fostering students’ imagination and reflection on possible ways to make concrete R&I processes more socially responsible.

Law and AI (Marta Infantino)

The course on ‘Law and AI’ focuses on how current and prospective legal rules constrain and govern data science and AI applications, especially in the European legal framework. The course will explore the general legal requirements currently applicable to data science and AI applications in Europe, as well as the ongoing debate on prospective legal measures that could in the future apply in the field. Through the analysis of regulatory texts and proposals, judicial decisions and scholarly studies, the course will in particular delve into the protection of privacy, obligations of transparency and fairness, and accountability rules.

Explainable AI (Luca Bortolussi)

The course on explainable artificial intelligence deals with the hot topic of identifying either efficient models that are also human comprehensible or to build explanations from highly efficient but black-box deep learning models. We will describe from a high level perspective the different approaches for deep learning models (e.g. local and global approximations/distillations and feature ranking) and also introduce some ideas from neuro-symbolic computing, attempting to combine logic and deep learning.