Supervised neural networks with memristor binary synapses. (22nd January 2018)

• Record Type: Journal Article
• Authors: Secco, Jacopo; Poggio, Mauro; Corinto, Fernando
• Other Names: Tetzlaff Ronald guestEditor.; Corinto Fernando guestEditor.; Picos Rogrigo guestEditor.; Ogorzalek Maciej guestEditor.
• Is Part Of: International journal of circuit theory and applications. Volume 46:Number 1(2018)
• Page Start: 221
• Page End: 233
• Publication Date: 2018-01-22
• Subjects: memristor -- spatio‐temporal patterns -- neural networks -- binary synapses -- pattern recognition -- memristor pulse programming; Electric circuit analysis -- Periodicals; 621.319205
• DOI: 10.1002/cta.2429
• Languages: English
• ISSNs: 0098-9886

Abstract

Summary: Memristors are emerging devices that promise the efficient implementation of synapses in artificial neural networks. Memristors have permitted the processing and analysis of a large amount of data in evolutionary learning artificial systems through signals that can be assimilated to human brain‐like neurotransmitters and synapses. In this manuscript, we present a memristor‐based neural network implementing the Stochastic Belief‐Propagation‐Inspired algorithm, an efficient supervised learning algorithm (which infers a classification rule from a set of labelled examples) suited for devices with very‐low‐precision synaptic weights. Synapses are represented by memristor devices described by the Generalized Boundary Condition Memristor model. We will thus demonstrate how to implement the key features of a machine learning algorithm in real‐world circuitry. Copyright © 2017 John Wiley & Sons, Ltd. Abstract : In this article, we present a memristor‐based neural network implementing the Stochastic Belief‐Propagation‐Inspired (SBPI) algorithm, an efficient supervised learning algorithm (which infers a classification rule from a set of labeled examples) suited for devices with very‐low‐precision synaptic weights. Synapses are represented by memristor devices described by the GBCM model.