The human body contains a group of strongly interrelated, delicate systems. Balance and stability are key to continuous function, and any perturbation to that balance can have dire consequences. When representing stability as a target class, many medical conditions and negative patient outcomes can be thought of as falling away from that target. This way we can model a patient's state using one-class classification to represent how far the patient is from "healthy", correlated to the likelihood of a systemic failure. In this talk I describe Hyperdimensional One-class classification (HD-OCC), a one-shot training algorithm for predicting systemic failure or identifying anomalies in a patient's data. I will discuss three different use case scenarios where HD-OCC was used. The first focuses on predicting future diagnosis of type 2 diabetes, the second experiment uses patient data to model sepsis and predict septic shock in patients within the intensive care unit, and finally, image processing by using pulmonary CT scans to detect patients with pulmonary diseases. The results show that, for these problems, HD-OCC outperforms other approaches to One-class classification and opens the way for fast an accurate models where data constraints hinder the application of deep learning approaches.
Neftali Watkinson is a PhD graduate from the University of California, Irvine. He is currently a postdoctoral researcher exploring practical applications of Hyperdimensional Computing. His passion for teaching and excellence in research has been recognized by the Fulbright fellowship, the Innovation in ICS award, the Miguel Velez fellowship award and the Latino Excellence and Achievement Award. He has participated in different projects, including a work of art displayed at the Venice Biennale 2019. Neftali's commitment to his students has been evident through his pedagogical research and multiple undergraduate research projects. He also co-designed an undergraduate course on practical learning on Edge Computing applications.
Faculty host: Marco Morales