Exponential convergence of adaptive importance sampling for Markov chains
adaptive proceduresbiased random walksexponential convergenceimportance samplinglearning algorithmsMarkov chainsmedical X-raysMonte Carlo simulationnuclear reactor operationsoil explorationparticle transportradiation dosage calculationszero-variance solution
Sums of independent random variables; random walks (60G50) Computational methods in Markov chains (60J22) Sampling theory, sample surveys (62D05) Monte Carlo methods (65C05) Stochastic particle methods (65C35) Numerical analysis or methods applied to Markov chains (65C40) Nuclear reactor theory; neutron transport (82D75) Potentials, prospecting (86A20) Biomedical imaging and signal processing (92C55)
The purpose of this paper is to establish conditions for exponential convergence of learning algorithms for Markov chains with scoring that arise in particle transport and have application to such fields as nuclear reactor operations, well logging for oil exploration and radiation dosage calculations for exposure to medical X-rays. Simulation of the `analog' chain is often so time consuming as to be impractical. Therefore, learning algorithms have been developed to simulate from approximate zero-variance important functions, and then use the results to improve the importance function. When the procedure is iterated, it has been shown numerically to converge exponentially to the solution. In this paper the problem is formally defined. The main result of this paper is a theorem stating the exponential convergence and its proof. The constraint (theorem assumptions) are shown to be rather mild to guarantee an exponential convergence for general state spaces. As an example, an idealized particle transport problem is treated by adaptive importance sampling. Finally, concluding remarks summarize the experience of the authors with the use of their approach.
- Adapative importance sampling on discrete Markov chains
- Adaptive Importance Sampling Technique for Markov Chains Using Stochastic Approximation
- ACCELERATED MONTE CARLO FOR PARTICLE DISPERSION
- Adaptive simulation using perfect control variates
- Adaptively learning an importance function using transport constrained Monte Carlo
- Rare Event Estimation for Computer Models
- Accelerating convergence in stochastic particle dispersion simulation codes
- Convergence of adaptive mixtures of importance sampling schemes
This page was built for publication: Exponential convergence of adaptive importance sampling for Markov chains
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4519099)