Path signatures provide a principled way to turn a time series or trajectory into a structured collection of features built from iterated integrals. Intuitively, these features capture not only increments, but also ordered interactions between channels across time (such as area-type effects and higher-order dependencies). This makes signatures particularly well-suited to sequential learning: they give a fixed-dimensional representation (after truncation) that is robust to irregular sampling, supports efficient computation via concatenation identities, and can serve as an expressive feature map for regression and classification—often enabling strong performance even with relatively simple downstream models.

Date: 16 March 2026
Place: BI Norwegian Business School, Room: A2-Blue-1  Format: 6 × 45-minute introductory lectures (three blocks of 2 × 45 minutes)
Audience: Open to anyone interested in mathematical models for time series, sequential data, and learning methods. No prior background in rough paths is assumed; we begin from first principles and build up toward computational aspects and current ML directions.

If you plan to attend, please feel free to drop by for the full day, or join for individual blocks depending on your interests. We hope this will provide a shared mathematical starting point for further discussions on sequential learning across the center. 

The event is free, but please register HERE

Program titles and abstracts: