Anzeige
Erschienen in:
2020 | OriginalPaper | Buchkapitel
Manifold Mixing for Stacked Regularization
verfasst von : João Pereira, Erik S. G. Stroes, Albert K. Groen, Aeilko H. Zwinderman, Evgeni Levin
Erschienen in: Machine Learning and Knowledge Discovery in Databases
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
In many real-world learning tasks, one has access to datasets consisting of multiple modalities, for example, various omics profiles of the patients coupled with medical records and other unstructured data sources. Often, the “core mechanism” (e.g. health or disease state) is reflected in all of these modalities and so this commonality can become more evident when the source domain (e.g. proteins) can accordingly transform the local geometry of the target (e.g. lipids). In this paper, we propose a novel algorithm that takes multiple data sources, constructs corresponding manifolds, and “mixes” information across them to find the common denominators in the observable outcomes. By leveraging manifold information from these different sources we obtain more robust and accurate results in comparison to standard methods. In the empirical evaluation on a clinical cohort related to ischaemia in patients with coronary artery disease, we demonstrate the applicability and efficacy of the proposed algorithm.