Print dated 7/10/2024 20:00

[IA/0205/EN] ARTIFICIAL INTELLIGENCE

General information
COMPUTER ENGINEERING, CYBERSECURITY AND ARTIFICIAL
Course
INTELLIGENCE
Curriculum PERCORSO UNICO
Course type MASTER DEGREE COURSE
Academic year 2024/2025
Year 1
Training activity type Compulsory subjects, characteristic of the class
Scope Computer Engineering
Language INGLESE
CFU 6 CFU
Didactic Activity
Practise, Lesson
Type
Exam type Oral exam
Evaluation Voto Finale
Teaching period First Semester (from 01/10/2024 to 28/02/2025)
Holders FUMERA GIORGIO
Teachers DEMONTIS AMBRA
Length 60 hours (12 hours Practise, 48 hours Lesson)
Frequency Mandatory
Didactic method Convenzionale
Subject area ING-INF/05
Location Cagliari - Università degli Studi
Main activity [IA/0114/EN] - INTELLIGENT SYSTEMS

Università degli studi di Cagliari Pag. 1 / 4

 Goals
This course is taught in English, and provides basic knowledge of some of the main approaches, methods,
and application fields of Artificial Intelligence, under the computer engineering perspective.

The learning outcomes expressed in terms of the Dublin descriptors are the following:

Knowledge and understanding

Students will know and understand some of the main approaches, models and algorithms of Artificial
Intelligence in the following fields: graph search; knowledge representation and reasoning using logical
languages, and using probabilistic methods; machine learning.

Applying knowledge and understanding

Students will be capable of solving simple problems in the above-mentioned fields, and of implementing and
testing computer programs to solve them.

Making judgements
Students will be able to analyze and design systems for solving simple problems in the above-mentioned
fields, taking into account the trade-off between computational complexity and solution quality, the latter in
terms of criteria such as accuracy and optimality.

Communication
Student will be capable of expressing and discussing issues related to Artificial Intelligence, to highlight
problems and to propose solutions.

Required skills
Elements of discrete mathematics (combinatorics), computer organization and architecture, at least one high-
level programming language.

Università degli studi di Cagliari Pag. 2 / 4

 Subjects
- Introduction and historical notes (1h)

- Graph search problems (lectures: 8h; exercises: 3h)

Formulation of search problems: state space, operators, goal, path cost, search tree. Uninformed search
strategies: breadth-first, depth-first, uniform cost; other search strategies (basics): depth-limited, iterative-
deepening depth-first, bidirectional. Informed, best-first search strategies: greedy search, A*; heuristic
functions. Computational complexity of search algorithms.

- Knowledge representation using logical languages (lectures: 6h; exercises: 3h)

Introduction: logical languages, inference. Propositional logic and predicate logic: syntax and semantics.
Inference in propositional logic: model checking, resolution, forward chaining and backward chaining
inference algorithms.

- Knowledge representation and reasoning under uncertainty (lectures: 7h; exercises: 2h)
Basic concepts of probability theory. Bayesian networks: structure and meaning, basic concepts on
probabilistic inference algorithms.

- Machine Learning: introduction and general concepts (lectures: 3h)

Machine Learning applications. Main Machine Learning paradigms. Supervised Learning. Generalization
capability. Overfitting.

- Decision Trees (lectures: 6h; exercises: 1h)

Decision tree structure. Learning algorithms: 'naive' algorithm and ID3. Pruning.

- Evaluating classifier performance (lectures: 1h)

Metrics and plots for performance evaluation: classification accuracy, confusion matrix, introduction to
ROC curve, precision and recall. Techniques for performance estimation: re-substitution, hold-out, cross-
validation.

- Neural networks (lectures: 8h; exercises: 1h, laboratory:3h ):

Historical notes . The perceptron; Rosenblatt's learning algorithm. Multilayer-Feedforward networks; the
backpropagation algorithm. Implementation of a neural network using the scikit-learn Python library.

- Ensemble methods (lectures: 4h)

Bias and variance of a classifier. Bagging, Random Forests, Boosting, AdaBoost.

- Exercises on machine learning (exercises: 3h)

Teaching methods
Lectures: 44 hours.
In-class exercises: 16 hours.
Lectures and in-class exercises will be held in person, possibly complemented with live streaming.

Università degli studi di Cagliari Pag. 3 / 4

 Verification of learning
The examination consists of a written test, possibly a supplemental oral test, if deemed necessary by the
instructors to better assess student's knowledge, and of a computer project.
The written and oral tests consist of open questions and exercises about all course topics.
The project can be developed individually or in groups of two. Its aim is to deepen the knowledge of one of
the course topics, through the implementation of suitable computer programs to solve specific problem
instances.
To pass the examination a pass mark in both the written test and the project is required; the final grade
(expressed in the numeric range 18-30) will be the sum of the two grades: up to 20 for the written test, up to
10 for the project.
To get a pass mark on the written/oral test students must possess a basic knowledge of all the course topics;
in particular, they must be capable of modelling simple problems in the fields considered in this course,
using suitable methods; must know the basic elements of the main approaches and algorithms to solve them;
must be able to apply such methods and algorithms to solve simple problem instances.
The grade of the written/oral test, expressed in a numeric scale from 12 to 20, depends on the degree of
knowledge of the approaches, methods and algorithms presented during the course, and on the degree of
complexity of the problems that the student will be able to model and to solve.
To get a pass mark on the project, students must possess: a sufficient mastery of related methods and
algorithms; the capability of developing computer programs to implement simple versions of such
algorithms; the capability of applying their programs to solve specific problem instances.
The grade of the project, expressed in a numeric scale from 6 to 10, depends on the degree of complexity of
the algorithms that the student will be capable of implementing, and on the capability of analyzing and
discussing the results of their application to specific problem instances.

Books
S. Russell and P. Norvig, "Artificial Intelligence - A Modern Approach", 4th Ed., Pearson, 2021, or previous
editions.

Extra info
Teaching material (lecture slides and exercises) is available on the web site https://unica-ai.github.io/

Università degli studi di Cagliari Pag. 4 / 4