Title: Singular decision problems in Health: A tutorial on principled discriminative machine methods
Abstract:
The huge majority of health decision problems are imbalanced –an alternative is much more frequent than the other– or have sample dependent costs –according to patient characteristics–, or both. These are two kinds of singular problem, those in which the conventional discriminative machine learning algorithms lead to poor results.
Many procedures to modify the learning processes have been proposed along the last three decades in order to improve the discriminative machine decisions for singular problems. However, most of them are purely empirical, and, consequently, they do not fully reduce the degradation risks.
This contribution reviews the recently proposed principled methods –based on Bayes theory– to deal with these kind of problems, discussing their theoretical pillars, and their practical design and implementation, emphasizing the advantages they provide.
Bio:
Doctor in Telecommunication Engineering, UPB, 1976. Full Professor, UPM, 1978. He served as a professor also at Univ. Santiago de Compostela and Universidad Carlos III de Madrid, where he is working at the present time. He has got an extensive experience at university administration positions (Department Head, School Dean, Deputy President).
His research interest focuses on digital signal processing and, mainly, on machine learning and its applications to communications, decision support systems, and inference from data. He has published more than one hundred journal papers in these subjects –most of them included in high impact index Journal Citation Report journals– and around three hundred conference papers. He has been a guest editor for three special issues of international journals, and a member of the organizing committee of dozens of international conferences. He has authored three books and edited another eight. He has supervised 33 doctoral dissertations. Ten of his Ph. D. students are now Full Professors at several universities.
With respect to knowledge transfer, he has been the principal investigator or director of more than 20 international and 70 national research projects and contracts. He also serves as a consultant or expert for many international and national organizations.
He introduced digital and statistical signal processing courses in Spain both at the undergraduate and graduate levels.
Prof. Figueiras-Vidal has received several prizes and awards along his academic life (National Graduation Award, UPC Doctorate Award, FUE and ESABE-CEOE Prizes to university-industry cooperation, UPM Foundation Prize to an Academic Life, Foundation “Technologies para la Defensa y la Seguridad” Prize, and “Miguel Catalán” Prize to his scientific trajectory, Madrid C., 2018). He is a Life Fellow of the IEEE, which awarded him with a Millennium Medal. He received the Doctor “Honoris Causa” degree from Vigo and San Pablo (Arequipa, Peru) universities. He is Correspondent Academician of Academia de Ingeniería de Mexico.
