用于可扩展训练和推理的稀疏深度神经网络分区

G. Demirci, H. Ferhatosmanoğlu
{"title":"用于可扩展训练和推理的稀疏深度神经网络分区","authors":"G. Demirci, H. Ferhatosmanoğlu","doi":"10.1145/3447818.3460372","DOIUrl":null,"url":null,"abstract":"The state-of-the-art deep neural networks (DNNs) have significant computational and data management requirements. The size of both training data and models continue to increase. Sparsification and pruning methods are shown to be effective in removing a large fraction of connections in DNNs. The resulting sparse networks present unique challenges to further improve the computational efficiency of training and inference in deep learning. Both the feedforward (inference) and backpropagation steps in stochastic gradient descent (SGD) algorithm for training sparse DNNs involve consecutive sparse matrix-vector multiplications (SpMVs). We first introduce a distributed-memory parallel SpMV-based solution for the SGD algorithm to improve its scalability. The parallelization approach is based on row-wise partitioning of weight matrices that represent neuron connections between consecutive layers. We then propose a novel hypergraph model for partitioning weight matrices to reduce the total communication volume and ensure computational load-balance among processors. Experiments performed on sparse DNNs demonstrate that the proposed solution is highly efficient and scalable. By utilizing the proposed matrix partitioning scheme, the performance of our solution is further improved significantly.","PeriodicalId":73273,"journal":{"name":"ICS ... : proceedings of the ... ACM International Conference on Supercomputing. International Conference on Supercomputing","volume":"29 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2021-04-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"7","resultStr":"{\"title\":\"Partitioning sparse deep neural networks for scalable training and inference\",\"authors\":\"G. Demirci, H. Ferhatosmanoğlu\",\"doi\":\"10.1145/3447818.3460372\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"The state-of-the-art deep neural networks (DNNs) have significant computational and data management requirements. The size of both training data and models continue to increase. Sparsification and pruning methods are shown to be effective in removing a large fraction of connections in DNNs. The resulting sparse networks present unique challenges to further improve the computational efficiency of training and inference in deep learning. Both the feedforward (inference) and backpropagation steps in stochastic gradient descent (SGD) algorithm for training sparse DNNs involve consecutive sparse matrix-vector multiplications (SpMVs). We first introduce a distributed-memory parallel SpMV-based solution for the SGD algorithm to improve its scalability. The parallelization approach is based on row-wise partitioning of weight matrices that represent neuron connections between consecutive layers. We then propose a novel hypergraph model for partitioning weight matrices to reduce the total communication volume and ensure computational load-balance among processors. Experiments performed on sparse DNNs demonstrate that the proposed solution is highly efficient and scalable. By utilizing the proposed matrix partitioning scheme, the performance of our solution is further improved significantly.\",\"PeriodicalId\":73273,\"journal\":{\"name\":\"ICS ... : proceedings of the ... ACM International Conference on Supercomputing. International Conference on Supercomputing\",\"volume\":\"29 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2021-04-23\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"7\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"ICS ... : proceedings of the ... ACM International Conference on Supercomputing. International Conference on Supercomputing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1145/3447818.3460372\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"ICS ... : proceedings of the ... ACM International Conference on Supercomputing. International Conference on Supercomputing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3447818.3460372","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 7

摘要

最先进的深度神经网络(dnn)具有显著的计算和数据管理要求。训练数据和模型的规模都在不断增加。稀疏化和修剪方法可以有效地去除dnn中的大部分连接。由此产生的稀疏网络对进一步提高深度学习中训练和推理的计算效率提出了独特的挑战。训练稀疏dnn的随机梯度下降(SGD)算法的前馈(推理)和反向传播步骤都涉及连续稀疏矩阵向量乘法(spmv)。我们首先为SGD算法引入了一个基于分布式内存并行spmv的解决方案,以提高其可伸缩性。并行化方法是基于表示连续层之间神经元连接的权重矩阵的逐行划分。然后,我们提出了一种新的超图模型来划分权重矩阵,以减少总通信量并确保处理器之间的计算负载平衡。在稀疏dnn上进行的实验表明,该方法具有较高的效率和可扩展性。利用所提出的矩阵划分方案,我们的解决方案的性能得到了进一步的显著提高。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
Partitioning sparse deep neural networks for scalable training and inference
The state-of-the-art deep neural networks (DNNs) have significant computational and data management requirements. The size of both training data and models continue to increase. Sparsification and pruning methods are shown to be effective in removing a large fraction of connections in DNNs. The resulting sparse networks present unique challenges to further improve the computational efficiency of training and inference in deep learning. Both the feedforward (inference) and backpropagation steps in stochastic gradient descent (SGD) algorithm for training sparse DNNs involve consecutive sparse matrix-vector multiplications (SpMVs). We first introduce a distributed-memory parallel SpMV-based solution for the SGD algorithm to improve its scalability. The parallelization approach is based on row-wise partitioning of weight matrices that represent neuron connections between consecutive layers. We then propose a novel hypergraph model for partitioning weight matrices to reduce the total communication volume and ensure computational load-balance among processors. Experiments performed on sparse DNNs demonstrate that the proposed solution is highly efficient and scalable. By utilizing the proposed matrix partitioning scheme, the performance of our solution is further improved significantly.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
期刊最新文献
Accelerating BWA-MEM Read Mapping on GPUs. Dynamic Memory Management in Massively Parallel Systems: A Case on GPUs. Priority Algorithms with Advice for Disjoint Path Allocation Problems From Data of Internet of Things to Domain Knowledge: A Case Study of Exploration in Smart Agriculture On Two Variants of Induced Matchings
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1