Benjamin Walker
Log Neural Controlled Differential Equations: Treating Time Series as Streams
Abstract
A stream represents a time series, by returning it's log-signature when queried over an interval. The Log-ODE method provides an efficient and accurate approximation to the solution of a controlled differential equation by treating the control as a stream. Log neural controlled differential equations (Log-NCDES) combine NCDEs with the Log-ODE method to achieve state-of-the-art performance on a range of multivariate time series classification benchmarks.
Our speaker
I am a PhD student in mathematics at the University of Oxford. The focus of my research is using the tools developed in the study of rough paths to improve machine learning methods for modelling multivariate time series. Additional areas of interest include modelling multimodal data in healthcare and developing a better understanding of Lip-gamma functions.
To become a member of the Rough Path Interest Group, register here for free.
