|
|||||
|
|
A dynamic programming approach to efficient sampling from Boltzmann distributions |
|
| Authors: | Archis Ghate Robert L. Smith |
| Affiliation: | a Industrial Engineering, University of Washington, Box 352650, Seattle, WA, 98195, USA b Industrial and Operations Engineering, University of Michigan, Ann Arbor, MI, 48109, USA |
| Abstract: | Markov chain methods for Boltzmann sampling work in phases with decreasing temperatures. The number of transitions in each phase crucially affects terminal state distribution. We employ dynamic programming to allocate iterations to phases to improve guarantees on sample quality. Numerical experiments on the Ising model are presented. |
| Keywords: | Simulated Annealing Markov chain Monte Carlo Cooling schedule Warm start |
| 本文献已被 ScienceDirect 等数据库收录！ | |