Jonathan Rosenthal, Shanchao Liang, Kevin Zhang, Lin Tan
{"title":"CaBaGe:使用 ClAss BAlanced Generator Ensemble 进行无数据模型提取","authors":"Jonathan Rosenthal, Shanchao Liang, Kevin Zhang, Lin Tan","doi":"arxiv-2409.10643","DOIUrl":null,"url":null,"abstract":"Machine Learning as a Service (MLaaS) is often provided as a pay-per-query,\nblack-box system to clients. Such a black-box approach not only hinders open\nreplication, validation, and interpretation of model results, but also makes it\nharder for white-hat researchers to identify vulnerabilities in the MLaaS\nsystems. Model extraction is a promising technique to address these challenges\nby reverse-engineering black-box models. Since training data is typically\nunavailable for MLaaS models, this paper focuses on the realistic version of\nit: data-free model extraction. We propose a data-free model extraction\napproach, CaBaGe, to achieve higher model extraction accuracy with a small\nnumber of queries. Our innovations include (1) a novel experience replay for\nfocusing on difficult training samples; (2) an ensemble of generators for\nsteadily producing diverse synthetic data; and (3) a selective filtering\nprocess for querying the victim model with harder, more balanced samples. In\naddition, we create a more realistic setting, for the first time, where the\nattacker has no knowledge of the number of classes in the victim training data,\nand create a solution to learn the number of classes on the fly. Our evaluation\nshows that CaBaGe outperforms existing techniques on seven datasets -- MNIST,\nFMNIST, SVHN, CIFAR-10, CIFAR-100, ImageNet-subset, and Tiny ImageNet -- with\nan accuracy improvement of the extracted models by up to 43.13%. Furthermore,\nthe number of queries required to extract a clone model matching the final\naccuracy of prior work is reduced by up to 75.7%.","PeriodicalId":501332,"journal":{"name":"arXiv - CS - Cryptography and Security","volume":"89 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"CaBaGe: Data-Free Model Extraction using ClAss BAlanced Generator Ensemble\",\"authors\":\"Jonathan Rosenthal, Shanchao Liang, Kevin Zhang, Lin Tan\",\"doi\":\"arxiv-2409.10643\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Machine Learning as a Service (MLaaS) is often provided as a pay-per-query,\\nblack-box system to clients. Such a black-box approach not only hinders open\\nreplication, validation, and interpretation of model results, but also makes it\\nharder for white-hat researchers to identify vulnerabilities in the MLaaS\\nsystems. Model extraction is a promising technique to address these challenges\\nby reverse-engineering black-box models. Since training data is typically\\nunavailable for MLaaS models, this paper focuses on the realistic version of\\nit: data-free model extraction. We propose a data-free model extraction\\napproach, CaBaGe, to achieve higher model extraction accuracy with a small\\nnumber of queries. Our innovations include (1) a novel experience replay for\\nfocusing on difficult training samples; (2) an ensemble of generators for\\nsteadily producing diverse synthetic data; and (3) a selective filtering\\nprocess for querying the victim model with harder, more balanced samples. In\\naddition, we create a more realistic setting, for the first time, where the\\nattacker has no knowledge of the number of classes in the victim training data,\\nand create a solution to learn the number of classes on the fly. Our evaluation\\nshows that CaBaGe outperforms existing techniques on seven datasets -- MNIST,\\nFMNIST, SVHN, CIFAR-10, CIFAR-100, ImageNet-subset, and Tiny ImageNet -- with\\nan accuracy improvement of the extracted models by up to 43.13%. Furthermore,\\nthe number of queries required to extract a clone model matching the final\\naccuracy of prior work is reduced by up to 75.7%.\",\"PeriodicalId\":501332,\"journal\":{\"name\":\"arXiv - CS - Cryptography and Security\",\"volume\":\"89 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-16\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Cryptography and Security\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.10643\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Cryptography and Security","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.10643","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
CaBaGe: Data-Free Model Extraction using ClAss BAlanced Generator Ensemble
Machine Learning as a Service (MLaaS) is often provided as a pay-per-query,
black-box system to clients. Such a black-box approach not only hinders open
replication, validation, and interpretation of model results, but also makes it
harder for white-hat researchers to identify vulnerabilities in the MLaaS
systems. Model extraction is a promising technique to address these challenges
by reverse-engineering black-box models. Since training data is typically
unavailable for MLaaS models, this paper focuses on the realistic version of
it: data-free model extraction. We propose a data-free model extraction
approach, CaBaGe, to achieve higher model extraction accuracy with a small
number of queries. Our innovations include (1) a novel experience replay for
focusing on difficult training samples; (2) an ensemble of generators for
steadily producing diverse synthetic data; and (3) a selective filtering
process for querying the victim model with harder, more balanced samples. In
addition, we create a more realistic setting, for the first time, where the
attacker has no knowledge of the number of classes in the victim training data,
and create a solution to learn the number of classes on the fly. Our evaluation
shows that CaBaGe outperforms existing techniques on seven datasets -- MNIST,
FMNIST, SVHN, CIFAR-10, CIFAR-100, ImageNet-subset, and Tiny ImageNet -- with
an accuracy improvement of the extracted models by up to 43.13%. Furthermore,
the number of queries required to extract a clone model matching the final
accuracy of prior work is reduced by up to 75.7%.