Adaptive importance sampling (AIS) algorithms are a rising methodology in signal processing, statistics, and machine learning. An effective adaptation of the proposals is key for the success of AIS. Recent works have shown that gradient information about the involved target density can greatly boost performance, but its applicability is restricted to differentiable targets. In this paper, we propose a proximal Newton adaptive importance sampler, an algorithm for estimating expectations with respect to non-smooth target distributions. We utilize a scaled Newton proximal gradient to adapt proposal distributions, obtaining efficient and optimized moves even when the target is not a differentiable density. We demonstrate the utility of the algorithm on two scenarios, either involving convex constraints or non-smooth sparse priors.
View on arXiv