The Neuro-Fuzzy Computing System With the Capacity of Implementation on a Memristor Crossbar and Optimization-Free Hardware Training

In this paper, first we present a new explanation for the relationship between logical circuits and artificial neural networks, logical circuits and fuzzy logic, and artificial neural networks and fuzzy inference systems. This shows us that neural networks are working in the same way as logical circ...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on fuzzy systems 2014-10, Vol.22 (5), p.1272-1287
Main Authors: Merrikh-Bayat, Farnood, Merrikh-Bayat, Farshad, Shouraki, Saeed Bagheri
Format: Article
Language:English
Subjects:
Artificial neural networks
Circuits
Computation
Computer simulation
Fuzzy logic
Hardware
Hebbian learning rule
Hebbian theory
Input variables
Logic gates
logical circuit
Memory devices
memristive device
memristor crossbar
Memristors
neural network
Neural networks
neuro-fuzzy computing system
Neurons
Training
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this paper, first we present a new explanation for the relationship between logical circuits and artificial neural networks, logical circuits and fuzzy logic, and artificial neural networks and fuzzy inference systems. This shows us that neural networks are working in the same way as logical circuits when the connection between them is through the fuzzy logic. However, the main difference between them is that logical circuits can be constructed without using any kind of optimization-based learning methods. Based on these results, we propose a new neuro-fuzzy computing system. As verified by simulation results, it can effectively be implemented on the memristor crossbar structure and therefore can be a good approach to emulating the computing power of human brain. One important feature of the designed hardware is that it has the potential to be directly trained using the Hebbian learning rule and without the need for any optimization. The system is also very capable of dealing with a large number of input-out training data without facing problems like overtraining or sensitivity to outliers.
ISSN:1063-6706
1941-0034
DOI:10.1109/TFUZZ.2013.2290140