{"title":"On accelerating stochastic neural networks","authors":"S. Ramakrishnan, D. Kudithipudi","doi":"10.1145/3109453.3123959","DOIUrl":null,"url":null,"abstract":"Stochastic computing for neural networks is gaining traction for energy efficiency in neuromorphic systems. Generally, the accuracy of these systems is correlated with the the stochastic bit stream length and requires long compute times. In this study we propose methods to accelerate a stochastic computing based feedforward neural network, extreme learning machine. A new stochastic training hardware unit for the extreme learning machine is also proposed. In the proposed design a performance boost of 60.61X is achieved for Orthopedic dataset with 212 bit stream length when tested on a Nvidia GeForce 1050 Ti. The design is also validated for two standardized datasets, an accuracy of 92.4% for MNIST dataset and 87.5% for orthopedic dataset is observed.","PeriodicalId":400141,"journal":{"name":"Proceedings of the 4th ACM International Conference on Nanoscale Computing and Communication","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2017-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the 4th ACM International Conference on Nanoscale Computing and Communication","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3109453.3123959","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Stochastic computing for neural networks is gaining traction for energy efficiency in neuromorphic systems. Generally, the accuracy of these systems is correlated with the the stochastic bit stream length and requires long compute times. In this study we propose methods to accelerate a stochastic computing based feedforward neural network, extreme learning machine. A new stochastic training hardware unit for the extreme learning machine is also proposed. In the proposed design a performance boost of 60.61X is achieved for Orthopedic dataset with 212 bit stream length when tested on a Nvidia GeForce 1050 Ti. The design is also validated for two standardized datasets, an accuracy of 92.4% for MNIST dataset and 87.5% for orthopedic dataset is observed.