Specifications

method with high level of performance. Currently BSVM has been numerically implemented with

three methods including Laplace approximation, mean field and variational method. This paper

proposes a new numerical method to implement BSVM with Hybrid Monte Carlo (HMC) sampling.

In the sampling process, an auxiliary variable was introduced to reduce the sampling difficulty,

and the frog leap was added to avoid the random walk. Experimental results on simulated data

regression indicate the proposed method is feasible. Simultaneously, the effect of parameters

such as step number, step size, and sampling number on performance of BSVM is analyzed.

Number of references: 14

Main heading: Support vector machines

Controlled terms: Monte Carlo methods

Uncontrolled terms: Auxiliary variables - Data regression - Effect of parameters

- Frog leap - Hybrid Monte Carlo - Laplace approximation - Mean field -

Probabilistic machines - Random Walk - Sampling process - Step number - Step

size - Variational methods

Classification code: 723 Computer Software, Data Handling and Applications - 922.2

Mathematical Statistics

DOI: 10.4156/AISS.vol5.issue1.35

Database: Compendex

20130208 新增 42 条

Accession number: 20130515975207

Title: Numerical test study on ground frost heave deformation caused by artificial

horizontal ground freezing method

Authors: Bi, Guiquan1, 2 ; Wang, Dong1 ; Li, Ning3/毕贵权;;李宁

Author affiliation:

1 School of Energy and Power Engineering, Lanzhou University of Technology, Lanzhou 730050,

China

2 State Key Laboratory of Frozen Soil Engineering, CAREERI, CAS, Lanzhou 730000, China

3 Institute of Geotechnical Engineering, Xi'an university of technology, Xi'an 710048, China

Corresponding author: Bi, G. (biguiquan@gmail.com)

Source title: Applied Mechanics and Materials

Abbreviated source title: Appl. Mech. Mater.

Volume: 256-259

Issue: PART 1

Monograph title: Advances in Civil Engineering II

Issue date: 2013

Publication year: 2013

Pages: 340-343

Language: English

ISSN: 16609336

E-ISSN: 16627482