Machine Learning

The subject introduces students to machine learning techniques applicable in environments that present restrictions in the distribution of the data used in the generation of the models: treatment of flows, incorporation of new experiences, evolution of concepts over time or the preservation of the privacy of the information. Their consideration requires specific training in the application of incremental learning techniques, detection of obsolescence and confidentiality in the handling of datasets.

1. To acquire knowledge of how the main incremental learning techniques work.

2. To apply incremental learning techniques for the analysis of real-time data in stationary and non-stationary environments.

3. To know the working principle of the main privacy-preserving learning paradigms.

The degree programme includes the following contents for this subject:
Real-time learning on continuous data (streaming data): incremental algorithms for supervised and unsupervised learning, learning models for the treatment of data obsolescence and concept changes in non-stationary data. Privacy-preserving learning paradigms (Privacy-by-default vs. Privacy-by-design).

The previous content will be developed through three main themes:

1. Machine Learning Online
2. Concept Drift
3. Federated Learning

The syllabus will be supplemented by a laboratory program, comprising two main blocks:
P1. Machine Learning Online and Concept Drift ( 8 HP)
P2. Federated Learning (3 HP)

Main bibliography
[1]. Bahri, M., Bifet, A., Gama, J., Gomes, H. M., & Maniu, S. (2021). Data stream analysis: Foundations, major tasks and tools.Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery,11(3), e1405.
[2]. Bifet, A., Gavalda, R., Holmes, G., & Pfahringer, B. (2018). Machine learning for data streams: with practical examples in MOA. MIT press
[3]. Gama, J., Žliobait;, I., Bifet, A., Pechenizkiy, M., & Bouchachia, A. (2014). A survey on concept drift adaptation.ACM computing surveys (CSUR),46(4), 1-37.
[4]. Gomes, H. M., Read, J., Bifet, A., Barddal, J. P., & Gama, J. (2019). Machine learning for streaming data: state of the art, challenges, and opportunities.ACM SIGKDD Explorations Newsletter,21(2), 6-22.
[5]. Hoi, S. C., Sahoo, D., Lu, J., & Zhao, P. (2021). Online learning: A comprehensive survey.Neurocomputing,459, 249-289.
[6]. Li, T., Sahu, A. K., Talwalkar, A., & Smith, V. (2020). Federated learning: Challenges, methods, and future directions. IEEE signal processing magazine, 37(3), 50-60. This syllabus will be interspersed with the internship program, which will consist of two main blocks:
[7]. Lu, J., Liu, A., Dong, F., Gu, F., Gama, J., & Zhang, G. (2018). Learning under concept drift: A review.IEEE Transactions on Knowledge and Data Engineering,31(12), 2346-2363.
[8]. Orabona, F. (2019). A modern introduction to online learning.arXivpreprint arXiv:1912.13213
[9]. Yang, Q., Liu, Y., Chen, T., & Tong, Y. (2019). Federated machine learning: Concept and applications. ACM Transactions on Intelligent Systems and Technology (TIST), 10(2), 1-19.


Complementary bibliography
[1]. AbdulRahman, S., Tout, H., Ould-Slimane, H., Mourad, A., Talhi, C., & Guizani, M. (2020). A survey on federated learning: The journey from centralized to distributed on-site learning and beyond. IEEE Internet of Things Journal, 8(7), 5476-5497.
[2] Bifet, A., Gavalda, R. "Learning from time-changing data with adaptive windowing." In Proceedings of the 2007 SIAM international conference on data mining, pp. 443-448. Society for Industrial and Applied Mathematics, 2007.
[3] Bifet, A., & Gavalda, R. (2009). Adaptive learning from evolving data streams. InAdvances in Intelligent Data Analysis VIII
[4] Bifet, A., Gavalda, R., Holmes, G., & Pfahringer, B. (2018). Machine learning for data streams: with practical examples in MOA. MIT press.
[5]. https://federated.withgoogle.com/
[6]. Gama, J., & Castillo, G. (2006). Learning with local drift detection. InAdvanced Data Mining and Applications: Second International Conference, ADMA 2006, Xi’an, China, August 14-16, 2006Proceedings 2(pp. 42-55). Springer Berlin Heidelberg
[7] Gama, J., Medas, P., Castillo, G., & Rodrigues, P. (2004, September). Learning with drift detection. InBrazilian symposium on artificial intelligence(pp. 286-295). Springer, Berlin, Heidelberg.
[8] Ghesmoune, M., Lebbah, M., & Azzag, H. (2016). State-of-the-art on clustering data streams. Big Data Analytics, 1, 1-27.
[9] Gomes, H. M., Montiel, J., Mastelini, S. M., Pfahringer, B., & Bifet, A. (2020, July). On ensemble techniques for data stream regression. In 2020 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.
[10] McMahan, B., Moore, E., Ramage, D., Hampson, S., & y Arcas, B. A. (2017, April). Communication-efficient learning of deep networks from decentralized data. In Artificial intelligence and statistics (pp. 1273-1282).
[11] Rahman, K. J., Ahmed, F., Akhter, N., Hasan, M., Amin, R., Aziz, K. E., ... & Islam, A. N. (2021). Challenges, applications and design aspects of federated learning: A survey.IEEE Access,9, 124682-124700.

The degree programme contemplates the following competences for this subject:

Basic and general competences
GC2 - To successfully tackle all the stages of an Artificial Intelligence project.
GC3 - To search for and to select the information needed to solve complex problems, handling with fluency the bibliographic sources in the field.
GC4 - To elaborate adequately and with some originality written compositions or motivated arguments, to write plans, work projects, scientific articles and to formulate reasonable hypotheses in the field.
GC5 - To work in teams, especially multidisciplinary teams, and to be skilled in time management, people and decision making.
CB6 - To possess and to understand knowledge that provides a basis or opportunity for originality in the development and/or the application of ideas, often in a research context.
CB7 - To apply their acquired knowledge and problem-solving skills in new or unfamiliar environments within broader (or multidisciplinary) contexts related to their field of study.
CB8 - To be able to integrate knowledge and face the complexity of making judgements based on incomplete or limited information, including reflections on the social and ethical responsibilities linked to the application of their knowledge and opinions.

Transversal competences
CT3 - To use the basic tools of information and communication technologies (ICT) necessary for the exercise of their profession and for lifelong learning.
TC7 - To develop the ability to work in interdisciplinary or transdisciplinary teams in order to offer proposals that contribute to sustainable environmental, economic, political and social development.
TC8 - To value the importance of research, innovation and technological development in the socio-economic and cultural progress of society.
CT9 - To have the ability to manage time and resources: develop plans, prioritise activities, identify critical ones, establish deadlines and meet them.

Specific competences
SC10.- To be able to construct, validate and apply a stochastic model of a real system based on the observed data and the critical analysis of the results obtained.
SC11.- To understand and master the main data analysis techniques and tools, both from a statistical and machine learning point of view, including those dedicated to the processing of large volumes of data, and the ability to select the most appropriate ones for problem solving.
CE12.- To be able to plan, formulate and solve all the stages of a data project, including the understanding and mastery of the fundamentals and basic techniques for the search and filtering of information in large data collections.
CE15.- To have knowledge of the computer tools in the field of automatic learning, and to be able to select the most appropriate one for the resolution of a problem.

The contents of the course will be taught indistinctly between lectures and interactive classes. The completion of all the proposed activities is necessary to pass the course.

Expository classes (theory): will consist of the explanation of the different sections of the course syllabus, with the help of electronic media (presentations, videos, etc.).

Interactive classes (practical): different practical problems related to the content of the subject will be posed for the student to solve individually or in groups.

Case studies: students may be presented with real or fictional work scenarios that present certain problems. Students will have to apply the theoretical and practical knowledge of the subject to find a solution to the question or questions posed. As a general rule, case studies will be carried out in groups. The different working groups will present and share their solutions.

Project-based learning: students may be given practical projects whose scope requires them to dedicate a significant part of their time to the subject.

Autonomous work: the scope and objectives of the projects, use cases and/or practical problems may require autonomous work on the part of the students, albeit under the supervision of the teaching staff.

Office hours: Office hours will be used to solve students' doubts related to the contents of the subject. These office hours can be both face-to-face and virtual (via email, virtual campus or Microsoft Teams platform).

Virtual Classroom: This subject will have a virtual classroom where students will be provided with all the necessary material in digital format. Different communication tools will also be provided to support both teaching and office hours, including videoconferencing, chat, e-mail, forums...

In order to pass the course, the student will have to carry out all the proposed activities and pass the corresponding exams.

First opportunity:
To pass the subject, the student must deliver and pass the proposed activities (50% of the final grade) and pass the final exam (50% of the grade).

Mid-term exams:
No mid-term exams will be held.


Second opportunity:
The grade obtained in the laboratory practices during the course is maintained, as well as its weight in the final grade. Students who have not reached the cut-off mark in the activities proposed during the previous call, may submit, prior to the second chance final exam, similar activities, which will be proposed by the teachers. Once both parts have been passed separately, the exam will account for the 50% of the final mark and the laboratory practices for the remaining 50%.

Exemption from attendance:
In case of dispensation of attendance, students will be examined under the same conditions as students in the first round.

Repeating students:
In case of repeating students, they will be examined under the same conditions as students in the first round.

No-show qualification:
The student will receive the qualification of "no-show" when he/she does not take the final exam.

Fraudulent performance of exercises or tests:
For cases of fraudulent performance of exercises or tests, the provisions of the "Normativa de avaliación do rendemento académico dos estudantes e de revisión de cualificacións" of the USC will apply.

Evaluation of competences:
In general, the development of the practical activities, projects and use cases, as well as the preparation of the theoretical topics will allow students to work on the basic, general and transversal competences of the subject. Specifically, through the projects and use cases, the competences CT7, CT9, CG5, CG4, CG2 will be assessed. The development of the practices, as well as the final test, will allow the evaluation of the specific competences: CE10, CE11, CE12, CE15.

The course has a workload of 3 ECTS. This figure leads to a workload of 75 hours, distributed as follows:
*Classroom work:
- Theoretical classes: 10 hours
- Practical classes: 7 hours
- Problem-based learning, seminars, case studies and projects: 4 hours
*Personal work of the student
- Theoretical classes: 10 hours
- Practical classes: 21 hours
- Problem-based learning, seminars, case studies and projects: 23 hours

Primary language: the subject will be taught in English.