Dynamic multiple predictive models for multi-view learning

• Grant number: 23/11704-0
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate (Direct)
• Start date: March 11, 2024
• End date: December 01, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Ana Carolina Lorena
• Grantee: Victor Castro Nacif de Faria
• Supervisor: Rafael Menelau Oliveira e Cruz
• Host Institution: Divisão de Ciência da Computação (IEC). Instituto Tecnológico de Aeronáutica (ITA). Ministério da Defesa (Brasil). São José dos Campos , SP, Brazil
• Institution abroad: Université du Québec à Montréal (UQÀM), Canada
• Associated to the scholarship: 22/10917-8 - On building ensembles of diverse and competent classifiers, BP.DD

Abstract

Combining classifiers and forming multiple predictive models or committees is a popular strategy to unite classification techniques of distinct biases and explore their complementary capabilities. In this project, we intend to investigate the generation of classifier committees for Multi-View Learning problems. Multi-View Learning is a Machine Learning paradigm in which diverse representations of the same data can be used in the learning process. For instance, in a medical dataset, there can be variables describing patients in the form of demographic, clinical, and laboratory exam-related information, among others. In each of these representations (views), different classification techniques can be more competent than others regarding predictive performance, justifying a dynamic selection of techniques. Furthermore, this selection can be global (for an entire view) or local, where one selects classifiers for each observation in a specific view. A meta-learning-based approach will be employed to select the classifiers to be combined. Herewith, the selection of algorithms suitable for new problems will be guided by knowledge on general properties of similar problems solved in the past. Professor Cruz's research group at ÉTS - Université du Québec has many works on the generation of classifier committees, with recent high impact papers published, justifying the choice of his research group for the internship. The work's supervisor in Brazil has experience with meta-learning, in particular, guided by measures that estimate the difficulty in solving classification problems. Therefore, there is a complementarity between the research groups that this collaboration will explore. (AU)