• Blended Learning & Flipped Classroom

Reality is different from its model, and from this difference uncertainty arises. When facing problems in the real world, dealing with uncertainty is needed. This course presents powerful modeling tools for engineers and in general people needing to model complex phenomena with Artificial Intelligence techniques. Among the application areas, we mention: (big) data analysis, classification, automatic control, robotics, modeling of artificial and natural phenomena, modeling of behaviors (e.g., of users and devices), decision support, forecasting. This course will introduce rigorously the theoretical fundamentals of different modeling approaches, will put in evidence their application possibilities, by comparing different models, examples, and application cases, and will introduce design techniques for systems based on these technologies. This course takes the place of the Soft Computing course, offered for may years, keeping the original motivation, dealing with uncertainty, while some of the specific techniques once presented in Soft Computing take now space appropriate for the development they had in the last years in other courses and have been migrated in Machine Learning, Neural Networks and Deep Learning, and Data Mining. In this Uncertainty for Artificial Intelligence course are treated more in details issues, techniques, and modeling approaches more strictly related to modeling the intrinsic uncertanty that is present in the

Goals for this course are listed in the following.

- Presentation of the main issues in modeling the real world, including the possible sources of uncertainty and how it is possible to reduce their effect or cope with them.

- Presentation  of basic knowledge about some technologies used to model uncertainty namely: Uncertainty measures, Logic-based approaches, Fuzzy Systems, Possibility systems, and Probabilistic Graphical Models (e.g., Bayesian Networks and Hidden Markov Models).

- Presentation of tools to implement the mentioned technologies.

- Analysis of paradigmatic case studies to understand the applicability issues of the mentioned technologies.

- Development of the ability to analyze a problem, to select the appropriate technology for a problem, to design data, architectures, and processes including the mentioned technologies

- Development of the ability to learn autonomously both declarative and procedural knowledge (thanks to innovative teaching methods, such as flipped class, and blended learning).

Acquisition of basic knowledge about some technologies to treat uncertainty in AI systems, namely: Uncertainty measures, Logic-based approaches, Fuzzy Systems, Possibility systems, and Probabilistic Graphical Models (e.g., Bayesian Networks and Hidden Markov Models). (DD1)

Acquisition of the ability to analyze a problem, to select the appropriate technology for a problem, to design data, architectures and processes for the mentioned technologies. (DD2, DD3)

Acquisition of basic operational abilities to implement the mentioned technologies. (DD2)

Acquisition of the ability to present both the knowledge, the process, and the proposed solutions, as well as to analyze results and data (DD4)

Acquisition of the ability to learn autonomously both declarative and procedural knowledge.

• Uncertainty sources that acffect models: typology, issues, and modeling approaches.
• Measure-based uncertainty modeling
• Logic-based uncertainty modeling
• Fuzzy models: fuzzy sets, fuzzy logic, fuzzy rules, motivations for fuzzy modeling, tools for fuzzy systems, design of fuzzy systems, applications.
• Bayesian networks: basics, design, learning, evaluation, applications
• Hidden Markov Models: basics, design, learning, evaluation, applications
• Applications: motivations, choices, models, case studies.

 No specific background is required.

The evaluation consists of both a written exam where both theoretical competence and modeling skills will be tested, and small team exercise(s), developed and evaluated during the course, intended to develop competence while following the course, so that the final written exam may be faced properly, with less concentrated effort. Attendance to lessons is important mainly to develop the modeling skills,  which cannot be acquired from books, but only from experience, and to participate to the development of team exercises in a way better than remote interaction, which is, however, still possible.

Students are strongly advised to consider ALL the resources, included books and not only the slides.