Fast Probabilistic Optimization from Noisy Gradients

Philipp Hennig

Abstract: Stochastic gradient descent remains popular in large-scale machine learning, on account of its very low computational cost and robustness to noise. However, gradient descent is only linearly effici...