Loading...

A CMOS implementation of simplified Linear Oscillatory Neuron (LON) model derived from FitzHugh - Nagumo model, application in artificial neural networks

Kashaninia, A. R ; Sharif University of Technology | 2006

307 Viewed
  1. Type of Document: Article
  2. Publisher: 2006
  3. Abstract:
  4. During this paper, a new simplified model is introduced for a neuron membrane, which is more or less, capable to mimic the dynamics of any specific physiological neuron membrane. This model is called Linear Oscillatory Neuron (LON) model, which is derived through special method of linearization applied to FitzHugh Nagumo[8] model in the neuron rest regime. As well, this linear model is terminated by a well known nonlinear system to achieve oscillatory and chaotic output, as it is observed in real neurons. Although some relatively exact models exist for special neurons, such as HH model [1,2] for giant axon of a squad (which is extracted through a Voltage clamp trial and curve fitting techniques), these models may not be applied to mimic every individual neurons. Moreover, those suffer of a lack of not involving basic alive neuron properties within themselves. It is shown that the resultant model is capable of exhibiting the addressed features of a biological neuron. These features are favorable when we want to use LON model in the nodes of an Artificial Neural Networks (ANNs) as they improve the performance of these networks in some cases e.g. involving inhibition property causes better competition of neurons which leads to better classification in pattern recognition applications[4]. A hardware implementation of LON model is proposed basing on gm - amplifiers and sub-threshold CMOS circuits. Finally, as an experiment, LONs are employed in an inhibitory network to handle segmentation of a MRI image of human head
  5. Keywords:
  6. CMOS integrated circuits ; Image segmentation ; Linear systems ; Mathematical models ; Nonlinear systems ; Pattern recognition ; State space methods ; Companding functions ; Demultiplication ; Generalized linear state space ; Linear Oscillatory Neuron ; State variables decoupling ; Volterra system ; Neural networks
  7. Source: WSEAS Transactions on Circuits and Systems ; Volume 5, Issue 6 , 2006 , Pages 863-871 ; 11092734 (ISSN)