Convergence of adaptive mixtures of importance sampling schemes
arXiv:0708.0711 · doi:10.1214/009053606000001154
Abstract
In the design of efficient simulation algorithms, one is often beset with a poor choice of proposal distributions. Although the performance of a given simulation kernel can clarify a posteriori how adequate this kernel is for the problem at hand, a permanent on-line modification of kernels causes concerns about the validity of the resulting algorithm. While the issue is most often intractable for MCMC algorithms, the equivalent version for importance sampling algorithms can be validated quite precisely. We derive sufficient convergence conditions for adaptive mixtures of population Monte Carlo algorithms and show that Rao--Blackwellized versions asymptotically achieve an optimum in terms of a Kullback divergence criterion, while more rudimentary versions do not benefit from repeated updating.
arXiv commentPublished at http://dx.doi.org/10.1214/009053606000001154 in the Annals of Statistics (http://www.imstat.org/aos/) by the Institute of Mathematical Statistics (http://www.imstat.org)
References in corpus (2)
Cited by in corpus (10)
- Adaptive approximate Bayesian computation
- Adaptive Importance Sampling in General Mixture Classes
- Generalized Multiple Importance Sampling
- Layered Adaptive Importance Sampling
- Improving Population Monte Carlo: Alternative Weighting and Resampling Schemes
- Bayesian Model Averaging in Astrophysics: A Review
- Hamiltonian Adaptive Importance Sampling
- Sequential Monte Carlo with transformations
- Rare Events via Cross-Entropy Population Monte Carlo
- Infinite-dimensional gradient-based descent for alpha-divergence minimisation