Bayesian sparse learning with preconditioned stochastic gradient MCMC and its applications

Yating Wang, Wei Deng, Guang Lin

Abstract: Highlights•Propose an adaptive Bayesian sparse deep learning algorithm for regression problems.•Optimize hyperparameters in priors and preconditioner via stochastic approximation.•SSGL prior ensures sparsity, results in less memory and computational resources use.•The preconditioner is scalable in practice and improves convergence speed.•Samples converge to the asymptotically correct distribution with a controllable bias.