BAMBI: blind accelerated multimodal Bayesian inference

Abstract

In this paper we present an algorithm for rapid Bayesian analysis that combines the benefits of nested sampling and artificial neural networks. The blind accelerated multimodal Bayesian inference (BAMBI) algorithm implements the MultiNest package for nested sampling as well as the training of an artificial neural network (NN) to learn the likelihood function. In the case of computationally expensive likelihoods, this allows the substitution of a much more rapid approximation in order to increase significantly the speed of the analysis. We begin by demonstrating, with a few toy examples, the ability of a NN to learn complicated likelihood surfaces. BAMBI's ability to decrease running time for Bayesian inference is then demonstrated in the context of estimating cosmological parameters from Wilkinson Microwave Anisotropy Probe and other observations. We show that valuable speed increases are achieved in addition to obtaining NNs trained on the likelihood functions for the different model and data combinations. These NNs can then be used for an even faster follow-up analysis using the same likelihood and different priors. This is a fully general algorithm that can be applied, without any pre-processing, to other problems with computationally expensive likelihood functions.Comment: 12 pages, 8 tables, 17 figures; accepted by MNRAS; v2 to reflect minor changes in published versio

Reconstructing the sky location of gravitational-wave detected compact binary systems: Methodology for testing and comparison

“…We have developed a set of sampling algorithms within the LSC Algorithms Library (LAL) [27], collected under LALINFERENCE [19], specifically for the analysis of gravitational-wave data, and for what is relevant here, coalescing binary signal models. The library contains two main stochastic parameterspace exploration techniques: Markov-Chain Monte Carlo (LALINFERENCE_MCMC [22]), and nested sampling (LALINFERENCE_NEST [24] and LALINFERENCE_BAMBI [28]). Different algorithms are included to validate results during the development stage and to explore a range of schemes to optimize the run time.…”

Training Deep and Recurrent Networks with Hessian-Free Optimization

“…HF continues to be useful in certain deep neural network applications Graff et al, 2012Graff et al, , 2014Boulanger-Lewandowski et al, 2012), and in particular those where highquality estimation of the gradient and curvature is practical (Kingsbury et al, 2012;Sainath et al, 2013b,c;Chung et al, 2014), or where the local curvature properties of the objective function are particularly extreme (e.g. when training RNNs on the long-term dependency problems of Hochreiter and Schmidhuber (1997)).…”

Parameter estimation for compact binaries with ground-based gravitational-wave observations using the LALInference software library

“…Well known stochastic sampling techniques -Markov chain Monte Carlo [13][14][15][16][17][18][19][20][21], Nested Sampling [22,23] and MultiNest/BAMBI [24][25][26][27] -have been used in recent years to develop algorithms for Bayesian inference on GW data aimed at studies of coalescing binaries. An underlying theme of this work has been the comparison of these sampling techniques and the cross-validation of results with independent algorithms.…”