Joint control variate for faster black-box variational inference

arXiv:2210.07290v4 Announce Type: replace
Abstract: Black-box variational inference performance is sometimes hindered by the use of gradient estimators with high variance. This variance comes from two sources of randomness: Data subsampling and Monte Carlo sampling. While existing control variates only address Monte Carlo noise, and incremental gradient methods typically only address data subsampling, we propose a new "joint" control variate that jointly reduces variance from both sources of noise. This significantly reduces gradient variance, leading to faster optimization in several …

abstract arxiv box control cs.lg data faster gradient incremental inference noise performance randomness sampling stat.ml type variance