EnhanceD Data stream Analysis : combining the signature method and machine learning algorithms – EDDA

Multidimensional time series are ubiquituous in many fields of science an industry. Most of the time, linear models are insufficient to capture the complex nature of the data.
Time series analysis from a machine learning point of view is largely driven by the desire to provide interpretable methods using nonlinear features.
The EDDA project intends to improve existing approaches by considering the iterated-integrals signature (IIS) features, which encodes nonlinear information in multivariate time series.
Strong interdisciplinary collaborations are needed to face the challenges of integrating in a relevant manner the IIS feature in machine learning algorithms.
The EDDA project aims at bringing together scientific experts from the disciplines of machine learning, stochastic analysis
and practitionners in oceanography to work together towards the common goals of 1) exploring and exploiting interpretability of the IIS feature, 2) identifying fundamental strategies and methods to elaborate IIS-enhanced machine learning algorithms, and 3) developing relevant applications in data-driven multidimensional time series analysis.