Deluceva:基于delta的快速视频分析神经网络推理

32nd International Conference on Scientific and Statistical Database Management Pub Date : 2020-07-07 DOI:10.1145/3400903.3400930

Jingjing Wang, M. Balazinska

{"title":"Deluceva:基于delta的快速视频分析神经网络推理","authors":"Jingjing Wang, M. Balazinska","doi":"10.1145/3400903.3400930","DOIUrl":null,"url":null,"abstract":"Modern video analytics requires efficient machine learning model serving and evaluation. We present Deluceva, a system that optimizes video applications by applying incremental and approximate computation techniques. Experiments on three real models and six videos show that our prototype system can achieve significant performance gains up to 79% with F1 errors below 0.1.","PeriodicalId":334018,"journal":{"name":"32nd International Conference on Scientific and Statistical Database Management","volume":"24 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2020-07-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"2","resultStr":"{\"title\":\"Deluceva: Delta-Based Neural Network Inference for Fast Video Analytics\",\"authors\":\"Jingjing Wang, M. Balazinska\",\"doi\":\"10.1145/3400903.3400930\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Modern video analytics requires efficient machine learning model serving and evaluation. We present Deluceva, a system that optimizes video applications by applying incremental and approximate computation techniques. Experiments on three real models and six videos show that our prototype system can achieve significant performance gains up to 79% with F1 errors below 0.1.\",\"PeriodicalId\":334018,\"journal\":{\"name\":\"32nd International Conference on Scientific and Statistical Database Management\",\"volume\":\"24 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2020-07-07\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"2\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"32nd International Conference on Scientific and Statistical Database Management\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1145/3400903.3400930\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"32nd International Conference on Scientific and Statistical Database Management","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3400903.3400930","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 2

摘要

现代视频分析需要高效的机器学习模型服务和评估。我们提出了Deluceva，一个通过应用增量和近似计算技术来优化视频应用的系统。在三个真实模型和六个视频上的实验表明，我们的原型系统可以获得高达79%的显着性能提升，F1误差低于0.1。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

查看原文本刊更多论文

Deluceva: Delta-Based Neural Network Inference for Fast Video Analytics

Modern video analytics requires efficient machine learning model serving and evaluation. We present Deluceva, a system that optimizes video applications by applying incremental and approximate computation techniques. Experiments on three real models and six videos show that our prototype system can achieve significant performance gains up to 79% with F1 errors below 0.1.

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

32nd International Conference on Scientific and Statistical Database Management

自引率

0.00%

发文量