Abstract:
Bayesian optimization is one of the most popular algorithms for optimization of expensive black-box functions within a limited time budget. We applied this technique in the setting of demographic inference, to the problem of inferring the history of populations’ development from genetic data sampled from present day individuals. Along the way, we used several non-orthodox tricks, like ensembling and log-transforms, improving over the vanilla Bayesian optimization. I will tell you about demographic inference, our approach and the results.