Machine Learning :classification and evaluation

Teacher(s)

Dupont Pierre;

Language

English
> French-friendly

Prerequisites

Main themes

• Learning as search, inductive bias
• Combinations of decisions
• Loss function minimization, gradient descent
• Performance assessment
• Instance-based learning
• Probabilistic learning
• Unsupervised classification

Learning outcomes

At the end of this learning unit, the student is able to :

Given the learning outcomes of the "Master in Computer Science and Engineering" program, this course contributes to the development, acquisition and evaluation of the following learning outcomes: INFO1.1-3 INFO2.3-4 INFO5.4, INFO5.5 INFO6.1, INFO6.5 Given the learning outcomes of the "Master [120] in Computer Science" program, this course contributes to the development, acquisition and evaluation of the following learning outcomes: SINF1.M4 SINF2.3-4 SINF5.3, SINF5.4 SINF6.1, SINF6.4 Students completing successfully this course will be able to: understand and apply standard techniques to build computer programs that automatically improve with experience, especially for classification problems; assess the quality of a learned model for a given task; assess the relative performance of several learning algorithms; justify the use of a particular learning algorithm given the nature of the data, the learning problem and a relevant performance measure; use, adapt and extend learning software. Students will have developed skills and operational methodology. In particular, they have developed their ability to: use the technical documentation to make efficient use of existing packages; communicate test results in a short report using graphics.

Content

• Decision Tree Learning: ID3, C4.5, CART, Random Forests
• Linear Discriminants: Perceptrons, Gradient-Descent and Least-Square Procedures
• Maximal Margin Hyperplanes and Support Vector Machines
• Deep Learning
• Probability and Statistics in Machine Learning
• Performance Assessment: Hypothesis testing, Comparing Learning Algorithms, ROC analysis
• Gaussian Classifiers, Fisher Linear Discriminants
• Bayesian Learning: ML, MAP, Optimal Classifier, Naive Bayes
• Instance-based learning: k-NN, LVQ

Teaching methods

• Lectures
• Computing projects including theoretical questions and practical applications. These projects are implemented in python. They are submitted and evaluated on the Inginious platform.

Evaluation methods

Computation of the global grade for the course

For the first session, the global grade for the course is solely based on the grades of the computing projects, submitted and evaluated during the semester.
This global grade is computed, right after the grading of the final project, as a weighted average of the project grades according to the following weighting scheme:

• project 1 = 10%
• project 2 = 15%
• project 3 = 10%
• project 4 = 15%
• project 5 = 50%

The projects are not evaluated again for the second session and may not be resubmitted. The grades for projects 1 to 4 are kept as such, while project 5 is replaced by a closed book written exam. This written exam is by default on paper or, when appropriate, on a computer. The global grade is computed according to the same weighting scheme used for the first session, with the written exam representing 50% of this global grade (and replacing the project 5 grade).

Rules for student collaboration and use of external resources, including generative AIs

Collaborative studying among students is encouraged through an exchange forum on Moodle. Each student is however expected to submit a personal solution to each project. The use of public resources (e.g. stackoverflow.com), including generative AIs (e.g. chatGPT) is permitted, as long as each (fragment of) code submitted by the student mentions specifically all the resources used.
The distribution or exchange between students of (fragments of) code is not authorized by any means (GitHub, Facebook, Discord, etc.), even after the project deadlines.
Failure to comply with these rules will be considered as plagiarism and/or cheating and will be sanctioned according to the Academic Regulations and Procedures.
These rules are explained in detail during the first class (see course Moodle site).
 

Other information

Prerequisites include :

• Introduction to probability (random variables, conditional probabilities, Bayes rule, ...) and statistics (confidence intervals, hypothesis tests, ...)
• Basics in linear algebra (matrix computation, linear systems, eigenvalues, ...) 
• Basics in calculus (e.g. partial derivatives from a multivariate function)
• Algorithmics (e.g. reasoning about the computational complexity of an algorithm)
• Full proficiency in Python programming

Online resources

moodle.uclouvain.be/course/view.php?id=1836

Bibliography

Des ouvrages complémentaires sont recommandés sur le site Moodle du cours.
Additional textbooks are recommended on the Moodle site for this course.

Teaching materials

• Les supports obligatoires sont constitués de l'ensemble des documents (transparents des cours magistraux, énoncés des travaux pratiques, compléments, ...) disponibles depuis le site Moodle du cours.
• Required teaching material include all documents (lecture slides, project assignments, complements, ...) available from the Moodle website for this course.

Faculty or entity

> INFO