----------------------------------------------------------------------------------------------------

     Department of Systems Engineering and Engineering Management

                    The Chinese University of Hong Kong

----------------------------------------------------------------------------------------------------

Date: Friday, September 18, 2026, 4:30pm to 5:30pm HKT

Venue: ERB 909, The Chinese University of Hong Kong

Title: Signature Approach for Contextual Bandits with Nonlinear and Path-dependent Rewards

Speaker: Dr. Xinyu Li, University of Oxford

Abstract:

 Real-world time series and sequential data are often non-stationary and nonlinear. At the same time, modern deep learning models often suffer from limited interpretability and, in principle, require a large amount of training data. In this talk, I will discuss how the signature transform can be exploited to address challenges arising from nonlinear and path-dependent sequential data and facilitate sequential decision-making.

In particular, we study contextual bandits when rewards depend nonlinearly on the history of observed contexts. Using the universal approximation property of signatures, we represent continuous path-dependent reward functionals by linear functionals of signature features. This allows us to develop DisSigUCB, a signature-based UCB algorithm that combines the expressive power of path-dependent models with the computational and theoretical advantages of linear contextual bandits. We establish sublinear regret guarantees, and prove consistency and asymptotic normality of the action-specific ridge estimators. Synthetic experiments and numerical applications to temperature sensor monitoring, sleep-stage classification, and hospital nurse staffing demonstrate the effectiveness of DisSigUCB relative to linear and kernelized contextual bandit baselines in nonlinear and path-dependent settings.

Biography:

Xinyu Li is a Departmental Lecturer in the Mathematical and Computational Finance Group at the University of Oxford. Before that, she was a Postdoctoral Research Associate in the Department of Mathematics at the University of Oxford, and a member of the Erlangen AI Hub.

She completed her Ph.D. in Industrial Engineering & Operations Research at UC Berkeley in Summer 2025, advised by Professor Xin Guo. Her research interests include game theory, stochastic control, and reinforcement learning, with applications to mathematical finance.

Everyone is welcome to attend the talk!

SEEM-5201 Website: http://seminar.se.cuhk.edu.hk