Machine Learning

To understand the main concepts associated with machine learning, as well as its mathematical basis. To analyse some of the most relevant and widely applied strategies in the automatic design of regressors and classifiers, as well as supervised, unsupervised and reinforcement learning approaches. In addition to analysing the pros and cons of the different approaches considered, some common problems that can arise from the training and test data sets used, both intrinsic and due to their inappropriate use, will be discussed.

1. Preliminary concepts of machine learning
2. Dimensionality reduction
3. Classification and linear regression
4. Decision trees
5. Committees
7. Neural networks and deep learning
6. Support Vector Machines
8. Clustering
9. Reinforcement learning

Basic:
Theodoridis: Machine learning: a Bayesian and optimization perspective, Academic Press, ISBN 9780128188033
C.M. Bishop. Pattern recognition and machine learning, Springer, 2006, ISBN: 978-0387310732
R.O. Duda, P.E. Hart, D.G. Stork. Pattern classification. Wiley Interscience, 2000. ISBN: 978-0471056690
R.S. Sutton, A.G. Barton. Reinforcement learning: an introduction. MIT Press, 2nd Edition, 2018. ISBN: ISBN: 978-0262039246

Complementary:
H. Daume. A course in machine learning. Autopublicado, 2017
P. Harrington. Machine learning in action. O'Reilly, 2012. ISBN 978-1617290183
J. Hurwitz, D. Kirsch. Machine learning for dummies. John Wiley & Sons, Inc., 2018. ISBN 9781119454953

CG8: Knowledge about basic technologies that allow to learn and develop new methods, with flexibility for adapting to new situations.

CG9: Ability to solve problems with iniciative, decision making, autonomy and creativity. Ability to communicate knowledge, abilities and skills of Computer Engineering.

TR1: Instrumentals: ability of analysis and synthesis, organization and planification. Oral and written communication in Galician, Spanish and English. Ability for information management. Problem solving. Decision making.

FB3: Ability to understand the basic concepts of discrete mathematics, logic, algorithmics and computational complexity, applied to solve problems in Computer Engineering.

RI5: Knowledge, administration and maintenance of systems, services and applications.

This subject uses software libraries that implement methods of learning from data that are able to adapt to new situations (competence CG8). Besides, there are processes of information extraction that require creative decision making and to communicate the information extracted from data (CG9) in oral and written forms, alongside with plannification of learning strategies and validation of the methods (TR1). These methods use concepts of discrete mathematics, and the analysis of their algorithmics and computational complexity are fundamental for their application to large volume data (FB3). Finally, the administration of systems and applications that implement the analyzed methods is also required (RI5).

Lectures (10 hours): master class type presentations to explain the contents of the different subjects, with special emphasis on the explanation and assimilation of concepts, mathematical foundations and the potential usefulness of machine learning.
Interactive classes (30 h): solving practical problems of classification, regression, clustering, dimensionality reduction and reinforcement learning.

Final exam with multiple-choice and short-answer questions on the contents covered in the lectures: 40% of the final mark.
Continuous assessment: evaluation of the performance and results achieved in the different practices of the subject: 60% of the final mark.

The submission of any of the practice reports (or any other evaluation of a practice) will mean that the student has opted to pass the subject. Therefore, from that moment on, even if he/she does not attend the final exam, he/she will have used up one opportunity.

If the student has to attend the second opportunity (July), the marks of any of the graded parts (final exam or continuous assessment, focused on the practicals) will be kept, as long as the mark is equal or higher than 5, otherwise both parts will have to be repeated. From then on, the same criteria will be applied to pass the subject as those already explained for the first opportunity.

In case of exam fraud the "Normativa de avaliación do rendemento académico d@s estudant@s e de revisión de cualificacións". will be applied.

Face-to-face work:
Blackboard lectures: 10h
Computer laboratory sessions: 30h
Assessment exams: 5h
Total: 45h

Personal work:
Autonomous study: 19h
Exercise solving: 7h
Computer programming: 28h
Exam preparation: 13h
Total: 67h

Class attendance and completion of the proposed practices using the machine learning libraries used in the course and through programming.

The support contents for the subject, both for the lectures and the interactive classes, will be available on the virtual campus of the subject.

The predominant language of instruction in the subject will be Galician.