Supervised machine learning

In supervised learning, intelligent systems learn by themselves methods for extracting information from data in a supervised manner, i.e. with known values for the quantities to be predicted automatically. Thus, intelligent systems learn to construct functions or models from annotated examples (training data). The learned functions have the capacity of generalization to be applied to new examples from which to make predictions or decisions. This course will teach the most important supervised learning methods and their basic applications (classification, prediction and regression). We will teach how to apply different techniques and choose the most appropriate one for each problem based on its characteristics, data volume and scalability. Some of the most relevant and widely applied strategies with various supervised learning approaches will be discussed. In addition to analyzing the pros and cons of the different approaches studied, some frequent problems that may arise from the training and test data set used, both intrinsic and due to an inappropriate use of the data, will be discussed. Thus, the main objective is to understand the concepts associated with machine learning from previously labeled data, as well as its mathematical foundations.

Unit 1. Preliminary concepts of machine learning: Regression and classification.

Unit 2. Model selection.

Unit 3. Nearest neighbor methods.

Unit 4. Rule-based systems.

Unit 5. Classification and linear regression.

Unit 6. Decision trees.

Unit 7. Support Vector Machines.

Unit 8. Ensembles: bagging, boosting and random forest.

1) Basic bibliography:

G. James, D. Witten, T. Hastie, R. Tibshirani. An Introduction to Statistical Learning with Applications in R. Ed. 1, New York: Springer, 2013. ISBN 978-1-4614-7137-0.

T. Hastie, R. Tibshirani, J. Friedman. The elements of statistical learning: data mining, inference, and prediction. Ed. 2, New York: Springer, 2009. ISBN 978-0-3878-4857-0.


2) Complementary bibliography:

C.M. Bishop. Pattern recognition and machine learning, Springer, 2006, ISBN: 978-0387310732

I.H. Witten, E. Frank, M.A. Hall, C.J. Pal, DATA MINNIG. Practical Machine Learning Tools and Techniques (Morgan Kaufmann Series in Data Management Systems). 4th Edition. ISBN-10:0128042915, ISBN-13:978-0128042915

P. Harrington. Machine learning in action. O'Reilly, 2012. ISBN 978-1617290183

A. Géron. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow. O'Reilly Media, Inc., ISBN:9781492032649, 2nd Edition

J. Hurwitz, D. Kirsch. Machine learning for dummies. John Wiley & Sons, Inc., 2018. ISBN 9781119454953

Throughout the course, students will develop the following skills:

BASIC:

CB2 - That students know how to apply their knowledge to their work or vocation in a professional manner and possess the competencies usually demonstrated through the development and defense of arguments and problem solving within their area of study.

GENERAL:

CG2.Ability to solve problems with initiative, decision-making, autonomy and creativity.

CG4. Ability to select and justify the appropriate methods and techniques to solve a specific problem, or to develop and propose new methods based on artificial intelligence.

GC5. Ability to design new computational systems and/or evaluate the performance of existing systems that integrate artificial intelligence models and techniques.

TRANSVERSAL:

TR1 - Ability to communicate and transmit their knowledge, skills and abilities.

TR2 - Ability to work in a team, in interdisciplinary environments and managing conflicts.

SPECIFIC:

CE1 - Ability to use mathematical and statistical concepts and methods to model and solve artificial intelligence problems.

CE12. Know the basics of algorithms and models of artificial intelligence for solving problems of certain complexity, understand their computational complexity and have the ability to design new models.

CE15. Know and know how to apply and explain correctly the validation techniques of artificial intelligence solutions.


Accordingly, the Learning Outcomes will be:
+ Know how to select the different supervised learning techniques to solve a problem in a certain domain.
+ Know the most important formal paradigms for supervised learning from data.
+ Know supervised learning techniques for classification and regression tasks.
+ Know the advantages and disadvantages of different supervised learning strategies, models and combination of models, but also evaluation metrics and validation methodologies to select the best models.
+ Know how to apply the algorithms and models studied in various use cases.

Lectures (30 h): master class presentations to explain the contents of the different topics, with special emphasis on the explanation and assimilation of concepts, mathematical foundations and the potential usefulness of supervised machine learning.

Interactive classes (20 h): resolution of practical classification and regression problems.

The course is 100% face-to-face.

Final exam with multiple-choice and short-answer questions on the contents covered in the lectures: 60% of the final grade.

Continuous evaluation: evaluation of the deliverables and results obtained in the different interactive sessions and exercises of the course: 40% of the final grade.

The attendance and participation of the students in the lectures as well as in the practicals, debates and/or seminars carried out during the course will be taken into account.

In order to pass the course it is necessary to pass both the final exam and the continuous evaluation.

The delivery of one of the internship reports (or any other evaluation of an internship) will mean that the student has chosen to take the course. Therefore, from that moment on, since he/she has not yet taken the final exam, one opportunity will be considered to have been used up.

In the second opportunity evaluation (July), the internship grade will be maintained if the student has a grade of 5 or more points. Otherwise, the student will have to be evaluated again for the practical part of the subject in this second opportunity. In any case, the student will have to take the final exam and the previous grade will not be retained. From then on, the same criteria already established for the first opportunity will be applied to pass the course.

In case of fraudulent examinations, the “Regulations for the evaluation of students' academic performance and grade review” (https://www.xunta.gal/dog/Publicados/2011/20110721/AnuncioG2018-190711-…) will be applied. Total or partial copying of any practice or theory exercise will automatically result in a grade of 0.0 in the subject and opportunity.

Classroom work time: 50h (total), divided into 30h (lectures), 20h (interactive classes).

Personal work time: 100h (total), including study, exercises and other evaluation activities.

It is recommended that students attend class and solve, verify and validate all the proposed exercises and practices (not only the evaluable ones), using the machine learning libraries used in the course and through direct programming.

Complementary means of teaching: virtual course on the platform provided by the USC, developed and constantly updated by the teachers of the subject.

The predominant teaching language is Spanish, but both in the bibliography and in the supplementary material there may be part of the contents in English.