Analog Learning Neural Network using Two-Stage Mode by Multiple and Sample Hold Circuits

Analog Learning Neural Network using Two-Stage Mode by Multiple and Sample Hold Circuits

Masashi Kawaguchi (Suzuka National College of Technology, Suzuka, Japan), Naohiro Ishii (Aichi Institute of Technology, Aichi, Japan) and Takashi Jimbo (Nagoya Institute of Technology, Aichi, Japan)
Copyright: © 2014 |Pages: 12
DOI: 10.4018/ijsi.2014010105
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

In the neural network field, many application models have been proposed. A neuro chip and an artificial retina chip are developed to comprise the neural network model and simulate the biomedical vision system. Previous analog neural network models were composed of the operational amplifier and fixed resistance. It is difficult to change the connection coefficient. In this study, we used analog electronic multiple and sample hold circuits. The connecting weights describe the input voltage. It is easy to change the connection coefficient. This model works only on analog electronic circuits. It can finish the learning process in a very short time and this model will enable more flexible learning.
Article Preview

Introduction

We propose the dynamic learning of the neural network by analog electronic circuits. This model will develop a new signal device with the analog neural electronic circuit. One of the targets of this research is the modeling of biomedical neural function. In the field of neural network, many application models have been proposed. And there are many hardware models that have been realized. These analog neural network models were composed of the operational amplifier and fixed resistance. It is difficult to change the connection coefficient.

Analog Neural Network

The analog neural network expresses the voltage, current or charge by a continuous quantity. The main merit is it can construct a continuous time system as well as a discrete time system by the clock operation. Obviously, the operation of the actual neuron cell utilizes analog. It is suitable to use an analog method for imitating the operation of an actual neuron cell. Many Artificial neural networks LSI were designed by the analog method. Many processing units can be installed on a single-chip, because each unit can be achieved with a small number of elements, addition, multiplication, and the nonlinear transformation. And it is possible to operate using the super parallel calculation. As a result, the high-speed offers an advantage compared to the digital neural network method (Mead, 1989; Chong, Salama, & Smith, 1992). In the pure analog circuit, the main problem is the achievement of an analog memory, how to memorize analog quantity (Lu & Shi, 2000). This problem has not been solved yet. The DRAM method memorizes in the capacitor as temporary memory, because it can be achieved in the general-purpose CMOS process (Saito & Inamura, 2003). However, when the data value keeps for a long term, digital memory will also be needed. In this case, D/A and A/D conversion causes an overhead problem. Other memorizing methods are the floatage gate type device, ferroelectric memory (FeRAM) and magnetic substance memories (MRAM) (Luthon & Dragomirescu, 1999; Yamada et al., 2000).

Pulsed Neural Network

Another hardware neural network model has been proposed. It uses a pulsed neural Network. Especially, when processing time series data, pulsed neural network model has good advantages. In particular, this network can keep the connecting weights after the learning process (Okuda, Doki, & Ishida, 2005). Moreover, the reason the learning circuit used the capacitor is that it takes a long time to work the circuits. In general, the pulse interval of the pulsed neural network is about 10μS. The pulsed neuron model represents the output value by the probability of neuron fires. For example, if the neuron is fired 50 times in a 100 pulse interval, the output value is 0.5 at this time. To represent the analog quantity using the Pulsed Neuron Model, it needs about 100 pulses. Thus, about 1mS is needed to represent the output analog signal on a pulsed neuron model.

In this study, we used the multiple circuits. The connecting weights describe the input voltage. It is easy to change the connection coefficient. This model works only on analog electronic circuits. It can finish the learning process in a very short time and this model will allow for more flexible learning. Recently, many researchers have focused on the semiconductor integration industry. Especially, low electrical power, low price, and large scale models are important. The neural network model explains the biomedical neural system. Neural network has flexible learning ability. Many researchers simulated the structure of the biomedical brain neuron using an electronic circuit and software.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 6: 4 Issues (2018): 1 Released, 3 Forthcoming
Volume 5: 4 Issues (2017)
Volume 4: 4 Issues (2016)
Volume 3: 4 Issues (2015)
Volume 2: 4 Issues (2014)
Volume 1: 4 Issues (2013)
View Complete Journal Contents Listing