On Parameter Adjustment of the Immune Inspired Machine Learning Algorithm AINE

Timmis, Jon (2000) On Parameter Adjustment of the Immune Inspired Machine Learning Algorithm AINE. Technical report. Kent University, Canterbury, Kent. CT2 7NF. (Full text available)

Download (288kB) Preview


A machine-learning algorithm based on the natural immune system metaphor has been developed, AINE (Artificial Immune Network). AINE developed from initial work on Artificial Immune Systems for data analysis, for which detailed experimentation was undertaken as to the affect of altering algorithm parameters had on the behaviour of the system. Two of the parameters, the network affinity threshold and mutation rate have been carried over into the new version, AINE. A third parameter the number of resources has been introduced into AINE as a means by which to control network size and create a stable network structure. This paper provides details of experiments, which alter these three parameters in AINE. It was expected that the two parameters taken from the AIS would, when altered, exhibit the same behaviour in AINE, that being the NAT scalar affecting network connectivity and mutation rate affecting network size and connectivity. Indeed, this was found to be the case. The third parameter, was designed to create stable network and previous work has shown this to be the case. It is shown in this paper, that the esource parameter can be used to control population size within the network.

Item Type: Monograph (Technical report)
Uncontrolled keywords: Artificial Immune System, machine learning, data analysis, SOFM
Subjects: Q Science > QA Mathematics (inc Computing science) > QA 76 Software, computer programming,
Divisions: Faculties > Sciences > School of Computing > Applied and Interdisciplinary Informatics Group
Depositing User: Mark Wheadon
Date Deposited: 13 Sep 2009 20:11 UTC
Last Modified: 26 Jun 2017 22:45 UTC
Resource URI: https://kar.kent.ac.uk/id/eprint/21937 (The current URI for this page, for reference purposes)
  • Depositors only (login required):