Adaptive Restructuring of Radial Basis Functions Using Integrate-and-Fire Neurons

Author

Marvel, Jeremy A.

Author_Institution

Nat. Inst. of Stand. & Technol., Gaithersburg, MD, USA

fYear

2014

fDate

3-6 Dec. 2014

Firstpage

189

Lastpage

194

Abstract

This paper proposes a neurobiology-based extension of integrate-and-fire models of Radial Basis Function Neural Networks (RBFNN) that adapts to novel stimuli by means of dynamic restructuring of the network´s structural parameters. The new architecture automatically balances synapses modulation, re-centers hidden Radial Basis Functions (RBFs), and stochastically shifts parameter-space decision planes to maintain homeostasis. Example results are provided throughout the paper to illustrate the effects of changes to the RBFNN model.

Keywords

neural net architecture; radial basis function networks; stochastic processes; RBFNN model; adaptive restructuring; dynamic restructuring; hidden radial basis function re-centers; homeostasis; integrate-and-fire models; integrate-and-fire neurons; network structural parameters; neurobiology-based extension; radial basis function neural networks; stochastic parameter-space decision planes; synapse modulation; Adaptation models; Biological neural networks; Biological system modeling; Neurons; Training; Vectors; feed-forward networks; machine learning; neural networks; radial basis functions;

fLanguage

English

Publisher

ieee

Conference_Titel

Machine Learning and Applications (ICMLA), 2014 13th International Conference on

Conference_Location

Detroit, MI

Type

conf

DOI

10.1109/ICMLA.2014.35

Filename

7033113