The use of mixture density networks in the emulation of complex epidemiological individual-based models

• DOI: 10.1371/journal.pcbi.1006869
• Publication Date: 2020-03-16T17:34:02Z
• Authors: Christopher N. Davis; T. Deirdre Hollin...; Quentin Caudron; Michael A. Irvine
• Keywords: Macro

Complex, highly-computational, individual-based models are abundant in epidemiology. For epidemics such as macro-parasitic diseases, detailed modelling of human behaviour and pathogen life-cycle required order to produce accurate results. This can often lead that computationally-expensive analyse perform model fitting, require many simulation runs build up sufficient statistics. Emulation provide a more computationally-efficient output the model, by approximating it using statistical model. Previous work has used Gaussian processes (GPs) achieve this, but these not deal with multi-modal, heavy-tailed, or discrete distributions. Here, we introduce concept mixture density network (MDN) its application emulation epidemiological models. MDNs incorporate both neural flexible tool for emulating variety outputs. We develop an MDN methodology demonstrate use on number simple incorporating normal, gamma beta distribution then explore stochastic SIR predict final size infection dynamics. have potential faithfully reproduce multiple outputs allow rapid analysis from range users. As such, open-access library method been released alongside this manuscript.