Corso di laurea magistrale in Informatica

The objectives of the course will be introduce students to the field of Machine Learning, that merges competencies of computer science (artificial intelligence, algorithms, optimization, databases) and statistics. The acquisition of Artificial Intelligence skills, of which machine learning is a cornerstone, is part of the specific training objectives of the CdS in Informatica (LM18), in particular among those of the address oriented to systems for information processing.

The course will teach the differences between tasks and models and will introduce the students to some of the popular models in Machine Learning. The program of the course will consist in the introductory part followed by a selection of the topics taken from the following list that will be agreed with the teachers on a single-student basis, according to the student's interests. The classification (into two classes and the transformation of a binary classification model into a multiple class model), concept learning by means of logical formulas, tree models and their purposes, hints to rule models and subgroup discovery, linear models (least squares, regression), perceptron, Support Vector Machines, Kernel methods. It will discuss the concept of bias-variance decomposition and the problem of overfitting. It will present the models based on distance, such as k-nearest neighbors, k-means clustering, hierarchical clustering, and density based clustering. The last part of the course discusses the probabilistic models (Naive Bayes, logistic regression and Expectation Maximization clustering), the evaluation of the models and of the statistical significance tests on the results. The laboratory part of the course will introduce the students to a practical open software suite ([Scikit Learn http://scikit-learn.org/stable/] in Python) that includes the algorithms of learning of the models seen during the course (and much more). With Scikit Learn we will approach some sessions of data analysis using some public available datasets.

1. KNOWLEDGE AND UNDERSTANDING of the main machine learning models.

2. KNOWLEDGE AND UNDERSTANDING of the basic statistical knowledge for data analysis, such as statistical significance tests, independence tests.
3. APPLYING KNOWLEDGE AND UNDERSTANDING of the modalities in which the data analysis process takes place, and in the use of a software suite for the application of Machine Learning principles and algorithms.
4. LEARNING SKILLS in the main Machine Learning models and on the models they produce.
5. LEARNING SKILLS in the software suites implementing Machine Learning algorithms and producing Machine Learning models.
6. MAKING JUDGEMENTS related to the choice of techniques to be adopted in each phase of the data analysis process.
7. MAKING JUDGEMENTS related to the choice of analysis technique to be adopted for a certain set of data.
8. APPLYING KNOWLEDGE AND UNDERSTANDING in the analysis of a machine learning algorithm.
9. APPLYING KNOWLEDGE AND UNDERSTANDING and COMMUNICATION SKILLS in the  interpretation and validation of analysis results and in communicating their meaning.

1. Introduction
2. Differences between tasks and models
3. Binary classification and transformation of a binary classification model into a multiple class model,
4. Concept learning by means of logical formulas,
5. Tree models and their purposes,
6. Hints on rule models,
7. Hints on subgroup discovery,
8. Linear models (least squares, regression), perceptron,
9. Support Vector Machines and Kernel methods.
10. Bias-variance decomposition
11. The problem of overfitting.
12. Models based on distance, such as k-nearest neighbors
13. K-means clustering
14. Hierarchical clustering,
15. Density based clustering.
16. Probabilistic models (naive Bayes, Expectation Maximization, logistic regression) 
17. Validation and verification of the results on data
18. Experiments with a real system on data sets: Scikit Learn.

Note: the order of the topics is indicative.

The lectures in class are done mainly with the aid of a computer (with slides). The execizes in laboratory sessions will be done on a computer with a software (Scikit Learn) or on Colab (the same software is available on remote on cloud). The same exercizes will be part of the oral examination. Teaching is in presence but we guarantee video recordings on the Moodle platform of the course. On the same platform some discussions will take place on discussion forums and on online meetings.

Oral interview in which the learning of the theoretical contents of the course and the experimentation on systems and algorithms is verified. The exams are held in presence. Here are listed by way of example some questions that were asked in previous years during the interview.

• Describe the main transformation operations between features
• Explain the goals of learning tree models
• Write pseudo-code for some classification or clustering algorithms
• Describe the measures and techniques for validating the classification process.
• Discuss the meaning and goals of learning concepts.
• Compare some algorithms for learning concepts against their ''bias''
• Describe and compare some of the approaches to clustering
• Illustrate one of the exercises carried out in the laboratory with Scikit Learn.

The laboratory exercizes with Scikit learn constitute a valid tool to the verification of the understanding of the theorical concepts and of their practice.