Presentation | 2021-02-12 A Defense Method for Machine Learning Poisoning Attacks in IoT Environments Considering the Removal Priority of Poisonous Data Tomoki Chiba, Yuichi Sei, Yasuyuki Tahara, Akihiko Ohsuga, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In recent years, machine learning has been attracting attention for its potential to further enrich people's lives. However, this has been accompanied by an increase in the number of vulnerabilities in systems that use machine learning. One such threat is the poisoning attack, which introduces poisonous data into the training data used to build machine learning models. The goal of this attack is to reduce the accuracy of the machine learning model or to output the prediction results that the attacker intended. In this paper, we propose a defense method to reduce the accuracy degradation of machine learning models caused by poisoning attacks. There are various scenarios for constructing machine learning models, but in this study, we assume an IoT environment, in which there are multiple sources of data, and an attacker may hide in one of them. In this study, we define a trust level for each source of data using poisonous data used in poisoning attacks, and remove data according to the trust level to suppress the accuracy degradation caused by poisoning attacks. In the evaluation experiments of the proposed method in this study, the detection accuracy of the proposed method is 80%, which is up to 50% higher than the accuracy of existing method. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | machine learning / security / IoT / poisoning |
Paper # | AI2020-36 |
Date of Issue | 2021-02-05 (AI) |
Conference Information | |
Committee | AI |
---|---|
Conference Date | 2021/2/12(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Online |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Naoki Fukuta(Shizuoka Univ.) |
Vice Chair | Yuichi Sei(Univ. of Electro-Comm.) / Yuko Sakurai(AIST) |
Secretary | Yuichi Sei(Nagoya Inst. of Tech.) / Yuko Sakurai(Tokyo Univ. of Agriculture and Technology) |
Assistant |
Paper Information | |
Registration To | Technical Committee on Artificial Intelligence and Knowledge-Based Processing |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Defense Method for Machine Learning Poisoning Attacks in IoT Environments Considering the Removal Priority of Poisonous Data |
Sub Title (in English) | |
Keyword(1) | machine learning |
Keyword(2) | security |
Keyword(3) | IoT |
Keyword(4) | poisoning |
1st Author's Name | Tomoki Chiba |
1st Author's Affiliation | University of Electro-Communications(UEC) |
2nd Author's Name | Yuichi Sei |
2nd Author's Affiliation | University of Electro-Communications(UEC) |
3rd Author's Name | Yasuyuki Tahara |
3rd Author's Affiliation | University of Electro-Communications(UEC) |
4th Author's Name | Akihiko Ohsuga |
4th Author's Affiliation | University of Electro-Communications(UEC) |
Date | 2021-02-12 |
Paper # | AI2020-36 |
Volume (vol) | vol.120 |
Number (no) | AI-362 |
Page | pp.pp.73-78(AI), |
#Pages | 6 |
Date of Issue | 2021-02-05 (AI) |