Ó³»­´«Ã½

Variational inference is becoming a widely used approach that enables scalable approximations of intractable posterior distributions. While sampling-based approaches such as MCMC provide a theoretically sound framework to generate samples from the true posterior distribution, these methods scale poorly with large data sets and models with many parameters. We demonstrate the use of variational inference to approximate the posterior parameter distribution of ordinary differential equation models applied to microbial community dynamics. We will present a scalable stochastic optimization approach to perform variational inference. Finally, we provide comparisons in parameter estimates and model predictions to alternative methods for Bayesian inference.