ENL_2017_BRISTOL.pdf (3.81 MB)
0/0

Recent Advances in Approximate Bayesian Computation Methodology (Application in structural dynamics).

Download (3.81 MB)
presentation
posted on 21.02.2017 by Anis Abdessalem, Keith Worden, Nikolaos Dervilis, David Wagg
"Recent advances in approximate Bayesian computation methodology: application in structural dynamics." Presentation given at the ENL Workshop, 9-10 January 2017, Bristol, United Kingdom.

In this work, a new variant of the approximate Bayesian computation (ABC) algorithms is presented based on the concept of the nested sampling (NS) algorithm proposed by Skilling [Ref.1] and an ellipsoidal sampling technique shown in Mukherjee et al. [Ref.2]. The ABC algorithms have been widely used for parameter estimation and model selection in different fields mainly when the likelihood function is intractable or cannot be approached in a closed form. However, those algorithms suffer from the high rejection rates during the sampling process. To overcome this problem, the authors propose a new variant of ABC algorithms called ABC-NS. The efficiency of the proposed algorithm will be compared with the ABC based on sequential Monte Carlo (ABC-SMC) sampler. The obtained results show how the ABC-NS outperforms the ABC-SMC in terms of computational efficiency while achieving the same or better results. Several examples have been proposed to illustrate the efficiency of the ABC-NS compared with the ABC-SMC for parameter estimation and model selection issues.

Ref.1: J. Skilling (2006) Nested sampling for general Bayesian computation. Bayesian Analysis, 1(4):833-860 [http://dx.doi.org/10.1214/06-BA127]
Ref.2: P. Mukherjee, D. Parkinson and A.R. Liddle (2006) A Nested Sampling Algorithm for Cosmological Model Selection. Astrophysical Journal Letters, 638 (2). L51-L54 [http://dx.doi.org/10.1086/501068]

Funding

EP/K003836/1

History

Licence

Exports