Title:
勾配を使用したニューラル・ネットワーク内のバックドアの検出
Document Type and Number:
Japanese Patent JP7374545
Kind Code:
B2
Abstract:
Mechanisms are provided for evaluating a trained machine learning model to determine whether the machine learning model has a backdoor trigger. The mechanisms process a test dataset to generate output classifications for the test dataset, and generate, for the test dataset, gradient data indicating a degree of change of elements within the test dataset based on the output generated by processing the test dataset. The mechanisms analyze the gradient data to identify a pattern of elements within the test dataset indicative of a backdoor trigger. The mechanisms generate, in response to the analysis identifying the pattern of elements indicative of a backdoor trigger, an output indicating the existence of the backdoor trigger in the trained machine learning model.
Inventors:
Lee, Taesun
Molloy, Ian, Michael
Carvalho, Wilka
Edwards, Benjamin, James
Chan, Jiaron
Chen, Bryant
Molloy, Ian, Michael
Carvalho, Wilka
Edwards, Benjamin, James
Chan, Jiaron
Chen, Bryant
Application Number:
JP2020554436A
Publication Date:
November 07, 2023
Filing Date:
April 10, 2019
Export Citation:
Assignee:
INTERNATIONAL BUSINESS MACHINES CORPORATION
International Classes:
G06T7/00; G06N3/04; G06N3/08; G06N20/00
Other References:
MUNOZ-GONZALEZ, Luis、他6名,Towards Poisoning of Deep Learning Algorithms with Back-gradient Optimization,Machine Learning,ARXIV.ORG,2017年08月29日,p.1-11,https://doi.org/10.48550/arXiv.1708.08689
CHEN, Xinyun、他4名,Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning,Cryptography and Security,ARXIV.ORG,2017年12月15日,p.1-18,https://doi.org/10.48550/arXiv.1712.05526
GU,Tianyu、他2名,BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain,Cryptography and Security,ARXIV.ORG[オンライン],2017年08月22日,p.1-13,インターネット:
CHEN, Xinyun、他4名,Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning,Cryptography and Security,ARXIV.ORG,2017年12月15日,p.1-18,https://doi.org/10.48550/arXiv.1712.05526
GU,Tianyu、他2名,BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain,Cryptography and Security,ARXIV.ORG[オンライン],2017年08月22日,p.1-13,インターネット:
Attorney, Agent or Firm:
Tadashi Taneichi