The subject is taught in the second semester of the 3rd year of the Artificial Intelligence degree. It is a natural continuation of the Data Mining subject (2nd year). It is also tightly linked to other subjects of the first semester, such as Machine Learning and Neural Networks, and Natural Language Processing.

The subject has a dual perspective. First, the theoretical roots of the machine learning paradigm are studied, and introduced a classifiers which naturally raises from these roots. Second, non-standard classification problems are studied and solved: that's, going beyond the "confort area" of supervised and unsupervised classification.

All these problems are crucial for the subjects in the 4th year of the degree.



In order to study this subject, the student needs to use the knowledge previously learned in the Data Mining subject, and other basic subjects such as Advanced Statistical Methods and Operation Research.

Output of the learning process:

- Knowledge on the theoretical-practical roots of the machine learning paradigm

1. Machine Learning fundamentals

- PAC learning, VC

- Loss functions

- Empirical Risk Minimization



2. Methods based on kernels

- Linear methods for classification - Support Vector Machines (SVMs)



3. Weakly supervised classification

- Semi-supervised classification

- Positive unlabeled learning

- Learning with label proportions

- Crowd learning

- Miscelanea: other types of weakly supervised problems



4. Non-standard supervised classification

- Multi-label learning

- Multi-dimensional classification

- Hierarchical classification

- Structured output prediction

Different teaching methodologies are used. Some of them, teaching classes with slides, presenting the theoretical parts and case studies of the subject. Some of these problems will be solved by the students.



On the other class, computer laboratories are used. Coding practices will be used to solve the exposed problems.



The student will receive feedback from the teacher, based on the output of these theoretical and laboratory classes. These works will be graded.

• Continuous Assessment System
• Final Assessment System
• Tools and qualification percentages:
  • Written test to be taken (%): 40
  • Realization of Practical Work (exercises, cases or problems) (%): 60

While continual and global evaluation systems are allowed, the first one, continual evaluation, is hardly recommended and will be the default system, as exposed in the UPV/EHU rules.



The student who, fulfilling the requirements to complete the continual evaluation, decides the global evaluation, needs to inform the teachers during the first 9 months of the course (in writing form).



The evaluation of the subject is continual, and composed of the following tests-exams:

1- Tests during the course: 60% of the final mark. This will be done by:

- Individual works to evaluate the concepts exposed during the classes.

- Group and individual works to evaluate how the students solve the problems exposed during the classes and practical sessions.



2- Writing exam in the official date fixed by the Faculty in the official exam-period: 40% of the final mark. It is a writing-exam, where the theoretical-practical concepts exposed in the subject will be evaluated.



The final mark is a sum of both, previous (sub)marks: but it is needed to obtain a minimum of 4 points (over 10) in each of the both tests-exams.



The student has the right to be evaluated by the "global evaluation system". To proceed with this, the student needs to present, in writing form, his/her election of the global system: during the first 9 weeks of the course. If the student does not present any work during the first 9 weeks of the course, it is understood that he/she opt for the global evaluation type.



In both types of evaluation, if the student is not present in the final writing exam, it is understood that he/she gives up, and will be qualified as "non-present".

The rules of the ordinary call will be applied.



The students who opted for the continual evaluation form can "save" for this extraordinary call the partial marks obtained in the tests and works done during the course. When these are completed, he/she only needs to perform the extraordinary exam of the subject, which covers the 40% of the final mark. It is also needed to obtain, at least, a 4 over 10, in each of the tests-exam, to be considered to pass the subject.

None.

Basic bibliography

Denis, F. (1998, October). PAC learning from positive statistical

queries. In International conference on algorithmic learning theory

(pp. 112-126). Springer, Berlin, Heidelberg.



Elkan, C., & Noto, K. (2008, August). Learning classifiers from only

positive and unlabeled data. In Proceedings of the 14th ACM SIGKDD

international conference on Knowledge discovery and data mining (pp.

213-220).



Hastie, T., Tibshirani, R., Friedman, J. H., & Friedman, J. H. (2009).

The elements of statistical learning: data mining, inference, and

prediction (Vol. 2, pp. 1-758). New York: springer.



Hernández-González, J., Inza, I., & Lozano, J. A. (2016). Weak

supervision and other non-standard classification problems: a

taxonomy. Pattern Recognition Letters, 69, 49-55.



Jaskie, K., & Spanias, A. (2019, July). Positive and unlabeled

learning algorithms and applications: A survey. In 2019 10th

International Conference on Information, Intelligence, Systems and

Applications (pp. 1-8). IEEE.





Soleimani, H., & Miller, D. J. (2017). Semisupervised, multilabel,

multi-instance learning for structured data. Neural computation,

29(4), 1053-1102



Van Engelen, J. E., & Hoos, H. H. (2020). A survey on semi-supervised

learning. Machine Learning, 109(2), 373-440.



Varma, P., Sala, F., He, A., Ratner, A., & Ré, C. (2019, May).

Learning dependency structures for weak supervision models. In

International Conference on Machine Learning (pp. 6418-6427). PMLR.





Zhou, Z. H. (2018). A brief introduction to weakly supervised

learning. National science review, 5(1), 44-53.

In-depth bibliography

None

Web addresses

- U. von Luxburg (2020). Statistical Machine Learning. YouTube videos serie. https://www.youtube.com/playlist?list=PL05umP7R6ij2XCvrRzLokX6EoHWaGA2cC