Hubert Wagner
Assistant Professor in Data Science (Department of Mathematics)
Areas of Interest
- Computational Geometry and Topology
- Machine Learning
- Algorithm Engineering
- Applications all of the above in various fields (astrophysics, biomedicine…)
Background
Before starting at UF, I did a postdoc at IST Austria with Herbert Edelsbrunner and a PhD at Jagiellonian University with Marian Mrozek. See my full CV .
Research
My work focuses on developing theory, algorithms and software for topological computations. The aim is to provide efficient tools useful in practical situations. The main streams of my work are:
- Imaging applications: Most of my work is in efficient computation of topological descriptors — especially persistent homology — which are used to analyze images. See my recent SoCG 23 paper which allows to handle 3D images with up to 8 Billions voxels. One interesting application of this type of computations is in analysis of cosmic microwave background radiation (paper in A&A).
- High-dimensional data: I use topological methods in conjunction with information theory to model higher-order interactions present in high-dimensional, non-Euclidean data. I’m particularly interested in datasets measured by Bregman divergences, which include the Kullback-Leibler divergence (aka relative entropy) widely used in machine learning. One interesting application is in detection of trojan attacks on deep neural networks (NeurIPS 2021 paper). This stream of work is supported by a 2022 Google Research Scholar award in Algorithms & Optimization.
Updated Publication List
Contact Information
Email: h{lastname}@ufl.edu
Office: 428 Little Hall (currently travelling)