Premium
A global search procedure for parameter estimation in neural spatial interaction modelling *
Author(s) -
Fischer Manfred M.,
HlaváčkováSchindler Kateřina,
Reismann Martin
Publication year - 1999
Publication title -
papers in regional science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.937
H-Index - 64
eISSN - 1435-5957
pISSN - 1056-8190
DOI - 10.1111/j.1435-5597.1999.tb00736.x
Subject(s) - maxima and minima , benchmark (surveying) , minification , robustness (evolution) , artificial neural network , computer science , mathematical optimization , conjugate gradient method , modal , backpropagation , trust region , estimation theory , machine learning , algorithm , mathematics , geography , mathematical analysis , biochemistry , chemistry , computer security , geodesy , polymer chemistry , radius , gene
. Parameter estimation is one of the central issues in neural spatial interaction modelling. Current practice is dominated by gradient based local minimization techniques. They find local minima efficiently and work best in unimodal minimization problems, but can get trapped in multimodal problems. Global search procedures provide an alternative optimization scheme that allows to escape from local minima. Differential evolution has been recently introduced as an efficient direct search method for optimizing real‐valued multi‐modal objective functions (Storn and Price 1997). The method is conceptually simple and attractive, but little is known about its behavior in real world applications. This article explores this method as an alternative to current practice for solving the parameter estimation task, and attempts to assess its robustness, measured in terms of in‐sample and out‐of‐sample performance. A benchmark comparison against backpropagation of conjugate gradients is based on Austrian interregional telecommunication traffic data.