Abstract
Two-terminal magnetoresistive random access memory (MRAM) devices provide a recent approach to intrinsically realizing stochastic neuronal behavior in cognitive architectures such as restricted Boltzmann machines (RBMs) for deep belief networks (DBNs). MRAM-based DBNs have achieved substantial energy and area improvements in comparison with the prior DBN hardware implementations. However, MRAM-based DBNs suffer from low accuracy in image classification. In order to resolve this problem, we present a new DBN-fuzzy system based on the combination of MRAM-based DBNs and fuzzy systems in the interest of improving the accuracy of MRAM-based neural networks. First, the MRAM-based DBN is employed to identify the top recognition results with the highest probability. Second, the fuzzy system is utilized to obtain the top-1 recognition results. We assess the accuracy of neural networks on the MNIST dataset, finding that the top-1 accuracy of the DBN-fuzzy neural network is improved from 64% to 82% in comparison to the MRAM-based DBNs for a $784\times10$ network.
Highlights
AND BACKGROUNDIn recent years, brain-inspired computing has investigated stochastic computing as the foundation of these frameworks
We implement the proposed deep belief networks (DBNs)-Fuzzy system for MNIST digit recognition dataset [34] including 60,000 training and 10,000 test sample images by employing the structure of a 784×10 DBN circuit in the Probabilistic Inference NetworkSimulator (PIN-Sim) framework with the device parameters that are listed in Table II and developing the TSK rule-based fuzzy system using Python scripts
We present an innovative image recognition technique for MNIST dataset on the basis of magnetoresistive random access memory (MRAM)-based DBNs and TSK rule-based fuzzy systems
Summary
Brain-inspired computing has investigated stochastic computing as the foundation of these frameworks. Stochastic computing models’ hardware implementation has not received adequate attention stochasticity is broadly being used in the computational neuroscience field. By contrast, emerging technologies such as spintronic devices demonstrate inherent randomness throughout their switching processes. These single-bit spin-based hardware units’ stochasticity behavior has been employed to realize efficient and compact neuronal hardware in diverse learning systems [1,2,3,4,5,6,7,8,9,10,11,12,13].
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
