Speaker: Dr. Martin Hohmann, Clinical Photonics Lab, Institute of Photonic Technologies of the University of Erlangen-Nuremberg (Germany)
Title: Statistics and machine learning – from classification to extracting unknowns
Location: Room ED38 (Building E, Faculty of Medicine, Campus Brabois Santé)
Abstract: This presentation explores the application of machine learning in various contexts, ranging from simple uses such as detecting carcinomas in the gastrointestinal tract to enabling new techniques for extracting physical features or unknown parameters. The challenges and solutions associated with machine learning are discussed, along with strategies for overcoming them, in the first part of the presentation. Additionally, scoring systems and manual alternatives are presented as means to accelerate research when machine learning is not feasible. While these methods represent a direct continuation of classical research, machine learning truly excels when combined with novel optical methodologies or customized neural networks, particularly when extracting parameters not present in the training or test data. The random laser and customized autoencoder network are highlighted as illustrative examples. Finally, Bayesian statistics is introduced as a complementary tool for deriving physical relationships, offering an alternative to machine learning.
Biography: Martin Hohmann is a postdoctoral researcher at the Institute of Photonic Technologies at the Friedrich-Alexander Universität Erlangen-Nürnberg. In 2021, he successfully defended his PhD thesis titled „Machine learning and hyper spectral imaging: multi spectral endoscopy in the gastro intestinal tract towards hyper spectral endoscopy“. Additionally, he serves as a mentor at Erlangen Graduate School in Advanced Optical Technologies (SAOT) and is an assistant editor at the Journal of Optics and Laser Technology (JOLT). His research and educational interests involve turbid media, with a focus on biophotonics and tissue optics. The central technologies of interest include hyperspectral imaging, random lasing, machine learning, and spectroscopic techniques. Throughout his career, he authored more than 40 articles.