Complements of Machine Learning

Information

Teachers: Ana Maria Tomé

Duration: One semester

Work hours: 162

Contact hours: 45

ECTS: 6

Scientific area: Computer Science

Objectives

Extend the students' knowledge of the machine learning field beyond the fundamental and more traditional principles.

Learning outcomes

Understand the formulation of different types of models.
Understand and apply probabilistic models.
Understand models based on latent variables.
Understand the various approaches to data transformation.
Understanding the implications of data dimensionality

Requirements

Knowledge of machine learning.

Grading

Grading will have three components: reading research papers that have already been published, and subsequently writing a synthesis (20%); practical project related to the techniques and / or a concrete application (30%); written exam that will assess the student's theoretical understanding (50%).

Methodology

Teaching will be based on theoretical classes to present the subject, accompanied by practical laboratory exercises and/or demos (notebooks).

Syllabus

Classic Classification Models
Design and analysis of learning experiments
Kernel Methods

The decision surface of a linear classifier
Definition of the optimal hyperplane; Definition of margin
The kernel functions and the kernel trick
Support Vector Machine classifier (SVM)
SVM for one class; Multiclass SVM

Probabilistic Methods

Estimation of discrete distribution parameters (Dirichlet and Beta)
Estimation of Gaussian distribution parameters
Bayesian estimation
Latent variable models and the EM algorithm; Estimation of a mixture model
Nonparametric Bayesian models
Latent Dirichlet models

New data representations

Intrinsic dimension of data. Transformations
Matrix factorization models: SVD; PCA; CCA; ICA; NMF
Dictionary learning (k-SVD)
Low-rank decomposition methods. Sparse models and robust PCA
Nonlinear models: kernel PCA and Local Linear Embedding (LLE)