DECIPHer - Methodology

A large part of the team effort focuses on methodological developments to address these limitations via a combination of data and models. A model, however useful, reliable and accurate, is complemented with data coming from additional physical or numerical experiments. The update from the prior model to the data-enhanced posterior model  is in theory "data-frugal" thanks to specifics of the physical systems we are interested in. These mechanical systems indeed often exhibit regularity, in a broad sense, including some degree of smoothness in space and time, symmetries and invariance (conserved quantities), sparsity and low-rankness in suitable representation variables and format, etc. An additional benefit is that data-driven methods scale with the effective dimension of the problem, as opposed to the ambient dimension, while naturally enjoying some robustness, a crucial aspect in practical applications.