{"title":"形成卷积神经网络训练/测试样本的实践方面","authors":"Y. Tomka, M. Talakh, V. Dvorzhak, O. Ushenko","doi":"10.31649/1681-7893-2022-43-1-24-35","DOIUrl":null,"url":null,"abstract":"The most common approaches to assessing the quality of training neural networks in the context of the problem of \"small training sets\" are analyzed. A review of the code implementation of the most universal approaches and ways of extending training/testing samples is carried out. The logic of the work of STN-module is analyzed. It can be inserted into existing convolutional architectures, giving neural networks the ability to actively spatially transform feature maps, conditional on the feature map itself, without any extra training supervision or modification to the optimization process.","PeriodicalId":142101,"journal":{"name":"Optoelectronic information-power technologies","volume":"161 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-12-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Practical Aspects of Forming Training/Test Samples for Convolutional Neural Networks\",\"authors\":\"Y. Tomka, M. Talakh, V. Dvorzhak, O. Ushenko\",\"doi\":\"10.31649/1681-7893-2022-43-1-24-35\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"The most common approaches to assessing the quality of training neural networks in the context of the problem of \\\"small training sets\\\" are analyzed. A review of the code implementation of the most universal approaches and ways of extending training/testing samples is carried out. The logic of the work of STN-module is analyzed. It can be inserted into existing convolutional architectures, giving neural networks the ability to actively spatially transform feature maps, conditional on the feature map itself, without any extra training supervision or modification to the optimization process.\",\"PeriodicalId\":142101,\"journal\":{\"name\":\"Optoelectronic information-power technologies\",\"volume\":\"161 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2022-12-28\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Optoelectronic information-power technologies\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.31649/1681-7893-2022-43-1-24-35\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Optoelectronic information-power technologies","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.31649/1681-7893-2022-43-1-24-35","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Practical Aspects of Forming Training/Test Samples for Convolutional Neural Networks
The most common approaches to assessing the quality of training neural networks in the context of the problem of "small training sets" are analyzed. A review of the code implementation of the most universal approaches and ways of extending training/testing samples is carried out. The logic of the work of STN-module is analyzed. It can be inserted into existing convolutional architectures, giving neural networks the ability to actively spatially transform feature maps, conditional on the feature map itself, without any extra training supervision or modification to the optimization process.