Efficient gPC-based quantification of probabilistic robustness for systems in neuroscience

TL;DR

This paper introduces a generalized polynomial chaos (gPC) framework as an efficient alternative to Monte Carlo methods for probabilistic robustness analysis in neuroscience models, enabling scalable uncertainty quantification.

Contribution

It develops and analyzes both intrusive and non-intrusive gPC approaches for scalable uncertainty quantification in complex neural models, addressing computational efficiency and accuracy.

Findings

gPC methods are scalable and faster than Monte Carlo

Trade-off between efficiency and accuracy of gPC approaches

Successful application to neural models with multiple dynamic regimes

Abstract

Robustness analysis is very important in biology and neuroscience, to unravel behavioural patterns of systems that are conserved despite large parametric uncertainties. To make studies of probabilistic robustness more efficient and scalable when addressing complex models in neuroscience, we propose an alternative to computationally expensive Monte Carlo (MC) methods by introducing and analysing the generalised polynomial chaos (gPC) framework for uncertainty quantification. We consider both intrusive and non-intrusive gPC approaches, which turn out to be scalable and allow for a fast comprehensive exploration of parameter spaces. Focusing on widely used models of neural dynamics as case studies, we explore the trade-off between efficiency and accuracy of gPC methods, and we adopt the proposed methodology to investigate parametric uncertainties in models that feature multiple dynamic regimes.