skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: On implementation of EM-type algorithms in the stochastic models for a matrix computing on GPU

Abstract

The paper discusses the main ideas of an implementation of EM-type algorithms for computing on the graphics processors and the application for the probabilistic models based on the Cox processes. An example of the GPU’s adapted MATLAB source code for the finite normal mixtures with the expectation-maximization matrix formulas is given. The testing of computational efficiency for GPU vs CPU is illustrated for the different sample sizes.

Authors:
 [1]
  1. Institute of Informatics Problems, Russian Academy of Sciences, Vavilova str., 44/2, Moscow, Russia MIREA, Faculty of Information Technology (Russian Federation)
Publication Date:
OSTI Identifier:
22391064
Resource Type:
Journal Article
Resource Relation:
Journal Name: AIP Conference Proceedings; Journal Volume: 1648; Journal Issue: 1; Conference: ICNAAM-2014: International Conference on Numerical Analysis and Applied Mathematics 2014, Rhodes (Greece), 22-28 Sep 2014; Other Information: (c) 2015 AIP Publishing LLC; Country of input: International Atomic Energy Agency (IAEA)
Country of Publication:
United States
Language:
English
Subject:
71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSICS; ALGORITHMS; EFFICIENCY; IMPLEMENTATION; M CODES; MATRICES; MIXTURES; PROBABILISTIC ESTIMATION; STOCHASTIC PROCESSES

Citation Formats

Gorshenin, Andrey K. On implementation of EM-type algorithms in the stochastic models for a matrix computing on GPU. United States: N. p., 2015. Web. doi:10.1063/1.4912512.
Gorshenin, Andrey K. On implementation of EM-type algorithms in the stochastic models for a matrix computing on GPU. United States. doi:10.1063/1.4912512.
Gorshenin, Andrey K. 2015. "On implementation of EM-type algorithms in the stochastic models for a matrix computing on GPU". United States. doi:10.1063/1.4912512.
@article{osti_22391064,
title = {On implementation of EM-type algorithms in the stochastic models for a matrix computing on GPU},
author = {Gorshenin, Andrey K.},
abstractNote = {The paper discusses the main ideas of an implementation of EM-type algorithms for computing on the graphics processors and the application for the probabilistic models based on the Cox processes. An example of the GPU’s adapted MATLAB source code for the finite normal mixtures with the expectation-maximization matrix formulas is given. The testing of computational efficiency for GPU vs CPU is illustrated for the different sample sizes.},
doi = {10.1063/1.4912512},
journal = {AIP Conference Proceedings},
number = 1,
volume = 1648,
place = {United States},
year = 2015,
month = 3
}
  • In this study, we present a stochastic method for the calculation of baryon 3-point functions which is an alternative to the typically used sequential method offering more versatility. We analyze the scaling of the error of the stochastically evaluated 3-point function with the lattice volume and find a favorable signal to noise ratio suggesting that the stochastic method can be extended to large volumes providing an efficient approach to compute hadronic matrix elements and form factors.
  • We analyze the parallel performance of randomized interpolative decomposition by de- composing low rank complex-valued Gaussian random matrices larger than 100 GB. We chose a Cray XMT supercomputer as it provides an almost ideal PRAM model permitting quick investigation of parallel algorithms without obfuscation from hardware idiosyncrasies. We obtain that on non-square matrices performance scales almost linearly with runtime about 100 times faster on 128 processors. We also verify that numerically discovered error bounds still hold on matrices two orders of magnitude larger than those previously tested.
  • The main difficulty in simulating diffusion processes at a molecular level in cell microdomains is due to the multiple scales involving nano- to micrometers. Few to many particles have to be simulated and simultaneously tracked while there are exploring a large portion of the space for binding small targets, such as buffers or active sites. Bridging the small and large spatial scales is achieved by rare events representing Brownian particles finding small targets and characterized by long-time distribution. These rare events are the bottleneck of numerical simulations. A naive stochastic simulation requires running many Brownian particles together, which is computationallymore » greedy and inefficient. Solving the associated partial differential equations is also difficult due to the time dependent boundary conditions, narrow passages and mixed boundary conditions at small windows. We present here two reduced modeling approaches for a fast computation of diffusing fluxes in microdomains. The first approach is based on a Markov mass-action law equations coupled to a Markov chain. The second is a Gillespie's method based on the narrow escape theory for coarse-graining the geometry of the domain into Poissonian rates. The main application concerns diffusion in cellular biology, where we compute as an example the distribution of arrival times of calcium ions to small hidden targets to trigger vesicular release.« less
  • We present detailed discussions on the stochastic Hamiltonians for noncritical string field theories on the basis of matrix models. Beginning from the simplest {ital c}=0 case, we derive the explicit forms of the Hamiltonians for the higher critical case {ital k}=3 (which corresponds to {ital c}={minus}22/5) and for the case {ital c}=1/2, directly from the double-scaled matrix models. In particular, for the two-matrix case, we do not put any restrictions on the spin configurations of the string fields. The properties of the resulting infinite algebras of Schwinger-Dyson operators associated with the Hamiltonians and the derivation of the Virasoro and {italmore » W}{sub 3} algebras therefrom are also investigated. Our results suggest a certain universal structure of the stochastic Hamiltonians, which might be useful for an attempt towards a background-independent string field theory. {copyright} {ital 1996 The American Physical Society.}« less