Metamodel-based dynamic algorithm configuration using artificial neural networks

We consider the problem of configuring algorithms dynamically by selecting algorithm parameter values adaptively. The research is motivated by the time dependency of system parameters throughout algorithm runtime in servicing systems: Depending on the customer arrival rate, switching algorithm param...

Full description

Saved in:
Bibliographic Details
Published in:International journal of general systems Vol. 53; no. 1; pp. 41 - 71
Main Authors: Dunke, Fabian, Nickel, Stefan
Format: Journal Article
Language:English
Published: Abingdon Taylor & Francis 02.01.2024
Taylor & Francis LLC
Subjects:
Algorithm configuration
Algorithms
artificial neural network
Artificial neural networks
Configurations
Customer services
Customers
dynamic decision making
Metamodels
Parameters
Performance prediction
simulation metamodeling
Switching theory
ISSN:0308-1079, 1563-5104
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We consider the problem of configuring algorithms dynamically by selecting algorithm parameter values adaptively. The research is motivated by the time dependency of system parameters throughout algorithm runtime in servicing systems: Depending on the customer arrival rate, switching algorithm parameters may be advisable to maintain quality of service. To this end, we develop a metamodel-based methodology for dynamic algorithm configuration: We first record algorithm performance under static system parameters. This knowledge is then translated into an artificial neural network (ANN) predicting performance for given system and algorithm parameters. The ANN finally serves as a metamodel determining optimal algorithm parameters dynamically when there is system parameter variation. Overall, the developed generic methodology for dynamic algorithm control facilitates a structured model-based approach to suitably respond to changing system conditions. The outline is adept to practical instantiation as demonstrated in two service systems where control parameters are adjusted adaptively to customer arrival rates.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0308-1079
1563-5104
DOI:10.1080/03081079.2023.2245124