The aim the course is to provide the required elements for understanding the foundations of the learning problem, with respect to the main algorithms for regression and classification. In particular, the course aims at enabling:
1. To gain basic notions, foundations, and methodologies of automatic learning, both supervised and unsupervised. 
2. To acquire adequate theroretical and applicative instruments typical of the field of machine learning. 
3. To achieve skills for applying tools and methodologies to different problems and scenarios. 
4. To obtain knowledge and competencies for using and adapting state of the art software systems for the solution of automatic learning problems.

01 Introduction
   01.01 Machine learning: motivations, problems, and introductory defintions
   01.02 Learning typologies

02 Mathematical optimization
   02.01 Zero order techniques
   02.02 First order techniques
   02.03 Second order techniques

03 Linear regression
   03.01 Least squares cost function
   03.02 Absolute deviation cost function
   03.03 Regression quality metrics
   03.04 Weighted regression

04 Binary linear classification
   04.01 Logistic regression
   04.02 Perceptron
   04.03 Support Vector Machines
   04.04 Binary classification quality metrics

05 Multi-class linear classification
   05.01 One-versus-All and multi-class classification
   05.02 Multi-class classification quality metrics
   05.03 Stochastic and mini-batch learning

06 Unsupervised learning
   06.01 Linear autoencoders
   06.02 Dimensionality reduction and principal components analysis
   06.03 Cluster analysis and the K-means algorithm 

07 Feature engineering, feature selection, nonlinear learning
   07.01 Feature engineering: histogram features, standard normalization
   07.02 Feature selection:  forward selection, regularization
   07.03 Nonlinear regression
   07.04 Nonlinear classification

08 Feature learning
   08.01 Universal approximators
   08.02 Generalization and overfitting
   08.03 Cross validation, regularization

09 Elements of neural networks
   09.01 Fully connected networks
   09.02 Activation functions
   09.03 Backpropagation algorithm

10 Laboratory activity
   10.01 Introduction to Python, Numpy, Scikit-learn
   10.02 Lab activity on linear regression
   10.03 Lab activity on linear classification: logistic regression, SVM
   10.04 Lab activity on unsuperivsed learning: PCA, K-means
   10.05 Lab activity on neural networks

 There are no mandatory prerequisites.

Knowledge and understanding: 

at the end of the course, the student will learn: the foundations to understand automatic learning problems, the design and analysis of the main supervised and unsupervised learning algorithms.

Applying knowledge and understanding:

the student will be able to understand the main features of the most common learning problems, and the related techniques for their solution. She/he will also be able to apply this knowledge for designing basic machine learning algorithms and the corresponding software systems.

Making judgements:

the student will be able to determine the adequacy and effectiveness of of machine learning algorithms and methodologies, particularly with respect to regression and classification problems.

Communication skills:

the student will learn to illustrate in a proper way the main learning techniques and to describe the factors affecting their effectiveness, with respect to specific application domains;  in particular, she/he will critically reason through a given mathematical descriptive language introduced during the course.

Learning skills:

at the end of the course, the student will gain a good autonomy level in critically understanding study materials regarding the main issues in machine learning, that will also enable to deal with novel design scenarios to which apply the achieved knowledge.

The teaching material prepared by the lecturer in addition to recommended textbooks (such as for instance slides, lecture notes, exercises, bibliography) and communications from the lecturer specific to the course can be found inside the Moodle platform › blended.uniurb.it

The teaching material prepared by the lecturer in addition to recommended textbooks (such as for instance slides, lecture notes, exercises, bibliography) and communications from the lecturer specific to the course can be found inside the Moodle platform › blended.uniurb.it

Teaching

Theory lectures and laboratory exercises.

Attendance

Although recommended, course attendance is not mandatory.

Course books

Jeremy Watt, Reza Borhani, Aggelos Katsaggelos: "Machine Learning Refined. Foundations, Algorithms, and Applications", Cambridge University Press, (2020).

Assessment

Learning achievements will be evaluated through a written exam (1 hour long, entailing multiple choice and open questions) and an oral exam.

Evaluation criteria for open questions are: knowledge level, articulation of answers (with the use of examples as well), adequacy level of the explanation, correct use of mathematical notation. Each criterion is evaluated on a four level scale (with equal weight assigned to each criterion). 

The oral exam, which can be taken only if the written exam has been passed (mark greater or equal than 18/30) consists of open questions about the course program. If passed, it determines an adjustement of the the previous mark, thus yielding the final mark. The oral exam is evaluated according to the following criteria: achieved knowledge, understanding of the basic principles of the subject, capability of presenting the topic in a rigorous manner. 

Disability and Specific Learning Disorders (SLD)

Students who have registered their disability certification or SLD certification with the Inclusion and Right to Study Office can request to use conceptual maps (for keywords) during exams.

To this end, it is necessary to send the maps, two weeks before the exam date, to the course instructor, who will verify their compliance with the university guidelines and may request modifications.