Citation

Lee, Gin Chong and Loo, Chu Kiong and Masuyama, Naoki (2019) Parameters Estimation in Topological Kernel Bayesian ART using Multi-objective Particle Swarm Optimization. In: 2018 IEEE Symposium Series on Computational Intelligence (SSCI), 18-21 Nov. 2018, Bangalore, India.

Text 77.pdf - Published Version Restricted to Repository staff only Download (179kB)

Abstract

Potentials in Topological Kernel Bayesian Adaptive Resonance Theory (TKBA) are advocated by the data specific parameters: kernel bandwidth σcim in correntropy induced metric (CIM) and kernel bandwidth σkbr in kernel Bayes' rule (KBR). This paper proposes a new calibration mechanism to implement Multi-objective Particle Swarm optimization (MOPSO) for parameters estimation in TKBA with the intention of searching for optimal values of parameters σcim and σkbr. Calibration mechanism is designed based on the measure of robustness, adaptability and the quality of the learned topological network. Two case studies has been empirically carried out using UCI real world dataset. Experiment results in case study I provide proof-of-concept of the proposed calibration mechanism. Case study II compares MOPSO calibrated TKBA with manual calibrated TKBA in terms of classification performance. Experiment results shows that MOPSO calibrated TKBA is able to enhance the capabilities of TKBA.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Particle swarm optimization
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	13 Jan 2022 02:22
Last Modified:	13 Jan 2022 02:22
URII:	http://shdl.mmu.edu.my/id/eprint/8965

Downloads

Edit (login required)