CRC/TRR 388/1: Statistical learning from path observations (B01)

Statistics and machine learning are now successfully applied to extremely complex and high-dimensional problems, often with data having a natural path-like structure, such as time series. This project aims to explore the statistical properties of the theory of rough paths and compare it with other methods. We will use path signatures as feature maps in learning, incorporating them into standard classification methods and analyzing the data with algebraic and geometric methods. The goal is to create explainable classifiers with theoretical guarantees for practical applications.