EasySurrogate: toolkit for creating surrogate models

Surrogate modelling toolkit, including neural-network based tools for the propagation of uncertainty through computational models with high-dimensional input spaces.

EasySurrogate is a toolkit designed to facilitate the creation of surrogate models, i.e. fast approximations, of (multiscale) simulations. It contains amongst others Deep Active Subspace (DAS) surrogates, which are neural networks combined with active subspace ideas to allow for dimension reduction. Possible applications include material science and drug discovery, where we are using DAS surrogates to look for low-dimensional structures in the high-dimensional input spaces found in these models. Other notable surrogate methods are quantized softmax networks, which are neural-network based bootstrapping methods that resample observed data, useful to create surrogate models for subgrid-scale terms in for instance turbulent flow simulations.

Like EasyVVUQ, the development of this software was funded by the EU Horizon 2020 Verified Exascale Computing for Multiscale Applications (VECMA) project (www.vecma.eu), and is currently maintained by CWI in collaboration with different partners from the UK-based SEAVEA project (https://www.seavea-project.org/).