Speaker: Anthony-Alexander Christidis

Topic: A New Ensemble Learning Framework for Modeling High-Dimensional Data

Date: Monday, September 11, 2023

Time: 3:00 – 4:00 PM Eastern Time (United States and Canada)

Virtual: Zoom link

Abstract: Ensemble methods have gained a lot of popularity in computational statistics and machine learning due to their flexibility and generally superior predictive performance compared to single-model methods. However, state-of-the-art ensemble methods rely on randomization and other forms of heuristics to generate diverse base learners, and are thus considered uninterpretable backbox algorithms. In this talk, we will introduce an entirely new framework for ensemble learning, free of randomization and heuristics. The models in these ensembles are learned by optimizing by a global objective function that generates diverse base learners while maintaining a degree of interpretability. Some aspects related to robustness will be discussed.