Event Detail

Event Type: 
Geometry-Topology Seminar
Monday, November 22, 2021 - 12:00 to 12:50
Kidd 280 or contact Christine Escher for a zoom link.

Speaker Info

Pacific Northwest National Laboratory

Presentation of topological frameworks for three machine learning tasks. First, we will discuss a novel approach for classification of aerial targets from motion track data using a combination of tools from dynamical systems together with persistent homology. As part of this effort we also present a new, widely applicable stability theorem relating time-series noise and distances between persistence images. Next, we will explore the notion of topology in language and describe insights gained from trying to understand the shape of text. Additionally, we will present a fascinating empirical study demonstrating the ability of topological features to strongly discriminate between valid and fraudulent research papers. Finally, we will conclude with a discussion of multi-modality fusion and initial results suggesting that topology provides a mathematically valid, universal feature space which can be leveraged for modality fusion.