B1748
Title: Bayesian deep neural networks: Optimality and adaptivity
Authors: Yongdai Kim - Seoul National University (Korea, South) [presenting]
Abstract: A Bayesian model for learning a certain sparse deep neural network is considered and an efficient MCMC algorithm is developed. In addition, we derive the posterior contraction rate for the proposed Bayesian model which is minimax optimal for various nonparametric regression models. Moreover, we prove that this optimality is adaptive to the unknown smoothness of the true function. By analyzing several benchmark data with our Bayesian model, we illustrate that the Bayesian model is superior to other nonparametric estimators.
