OVERVIEW

Machine learning is establishing as a very interesting scientific area thanks to the availability of more and more powerful computers and of algorithms allowing its application to the most diverse sectors, including key sectors for economic and societal development. This course presents the main machine learning methodologies aimed to “pattern recognition”, in particular for the classification of structurally represented data.

AIMS AND CONTENT

LEARNING OUTCOMES

In this module several Machine learning methods applied to pattern recognition are presented and - in particular - the case of classification of images from real domains is discussed. Students will learn how to classify samples on the basis of decision theory and linear and nonlinear classifiers (MDM, k-nn, SVMs, Random forest). Neural Networks and Deep Learning • Supervised Probability Density Estimation • Feature Reduction • Accuracy of Supervised Classifiers • Unsupervised Classifiers (Clustering) • Fuzzy Unsupervised Classifiers (fuzzy c-means).

AIMS AND LEARNING OUTCOMES

Purpose of this course is to provide the fundamentals of machine learning and to present some advanced methods with special reference to the classification of structurally represented data; in addition, examples will be presented and applications will be discussed related to signal and image classification. Students will learn to represent the features (measures, attributes, characteristics) of a set of samples to be classified by  a multidimensional vector space; will be able to reduce the dimensionality of such a representation limiting the information loss, to estimate the probabilistic distributions of data, to classify samples with classical techniques and with more recent ones (SVMs, neural networks, deep learning, classifier ensembles), evaluate or estimate the accuracy of a supervised classifier, extract the natural classes (clusters) present in a data set, also representing the related uncertainty with the “fuzzy” approach and validating the related results.

PREREQUISITES

Calculus (functions of one or more variables, integrals, functional optimization); probability theory and random variables; matrix calculus.

TEACHING METHODS

Class lessons on theory, applications, and problem solutions. Lab exercises based on guided software implementation of some of the techniques learned at lesson.

SYLLABUS/CONTENT

In this course several Machine learning methods applied to pattern recognition are presented and their application to images from real domains are discussed. In particular, the following methods are considered:

• representation of the features of a set of samples to be classified by a multidimensional vector space
• Decision Theory
• Supervised Probability Density Estimation
• Dimensionality Reduction of the Feature Space
• Neural Networks and Deep Learning
• Other Linear and Nonlinear Classifiers 
• Accuracy of Supervised Classifiers
• Unsupervised Classifiers (Clustering)
• Fuzzy Sets and Clustering.

RECOMMENDED READING/BIBLIOGRAPHY

Slides presented at lesson will be available in Aulaweb.

Recommended books:

• Duda, R. O., Hart, P. E., Stork, D. G.: 2001, Pattern classification, 2nd ed., Wiley.
• Fukunaga, K.: 1990, Introduction to statistical pattern recognition, 2a ed., Academic Press.
• Goodfellow I., Bengio Y., and Courville A., Deep learning, MIT Press, 2016.

More specific references are included in each of the chapters in which the slides are grouped. 

TEACHERS AND EXAM BOARD

Exam Board

SEBASTIANO SERPICO (President)

SILVANA DELLEPIANE

IGNACIO GASTON MASARI

MARTINA PASTORINO

GABRIELE MOSER (President Substitute)

LESSONS

LESSONS START

https://corsi.unige.it/10378/p/studenti-orario

Class schedule

The timetable for this course is available here: Portale EasyAcademy

EXAMS

EXAM DESCRIPTION

Mandatory written examination about the topics in the syllabus of the course, with maximum admissible mark equal to 24/30. If a student obtains a sufficient mark in this written exam, then he/she can also optionally take an additional oral examination with maximum admissible mark equal to 30/30 with honors.

ASSESSMENT METHODS

The assessment is based on questions aimed at verifying the understanding of the concepts taught, the ability to choose the most suitable methodology in the presence of specific cases and to apply the algorithms presented in class and during the exercises to simple problems.

Exam schedule

FURTHER INFORMATION

Students who have valid certification of physical or learning disabilities on file with the University and who wish to discuss possible accommodations or other circumstances regarding lectures, coursework and exams, should speak both with the instructor and with Professor Federico Scarpa (federico.scarpa@unige.it ), the School's disability liaison.