Seq2Seq Surrogates of Epidemic Models to Facilitate Bayesian Inference

Epidemic models are powerful tools in understanding infectious disease. However, as they increase size and complexity, can quickly become computationally intractable. Recent progress modelling methodology has shown that surrogate be used to emulate complex epidemic with a high-dimensional parameter space. We show deep sequence-to-sequence (seq2seq) serve accurate surrogates for sequence based model parameters, effectively replicating seasonal long-term transmission dynamics. Once trained, our predict scenarios several thousand times faster than the original model, making them ideal policy exploration. demonstrate replacing traditional learned simulator facilitates robust Bayesian inference.