A federated learning scheme meets dynamic differential privacy
Abstract Federated learning is a widely used distributed learning approach in recent years, however, despite model training from collecting data become to gathering parameters, privacy violations may occur when publishing and sharing models. A dynamic approach is proposed to add Gaussian noise more...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2023-09-01
|
| Series: | CAAI Transactions on Intelligence Technology |
| Subjects: | |
| Online Access: | https://doi.org/10.1049/cit2.12187 |
| _version_ | 1797682976120635392 |
|---|---|
| author | Shengnan Guo Xibin Wang Shigong Long Hai Liu Liu Hai Toong Hai Sam |
| author_facet | Shengnan Guo Xibin Wang Shigong Long Hai Liu Liu Hai Toong Hai Sam |
| author_sort | Shengnan Guo |
| collection | DOAJ |
| description | Abstract Federated learning is a widely used distributed learning approach in recent years, however, despite model training from collecting data become to gathering parameters, privacy violations may occur when publishing and sharing models. A dynamic approach is proposed to add Gaussian noise more effectively and apply differential privacy to federal deep learning. Concretely, it is abandoning the traditional way of equally distributing the privacy budget ϵ and adjusting the privacy budget to accommodate gradient descent federation learning dynamically, where the parameters depend on computation derived to avoid the impact on the algorithm that hyperparameters are created manually. It also incorporates adaptive threshold cropping to control the sensitivity, and finally, moments accountant is used to counting the ϵ consumed on the privacy‐preserving, and learning is stopped only if the ϵtotal by clients setting is reached, this allows the privacy budget to be adequately explored for model training. The experimental results on real datasets show that the method training has almost the same effect as the model learning of non‐privacy, which is significantly better than the differential privacy method used by TensorFlow. |
| first_indexed | 2024-03-12T00:07:43Z |
| format | Article |
| id | doaj.art-feeb1f2a28144cef86de0f23547d9aa5 |
| institution | Directory Open Access Journal |
| issn | 2468-2322 |
| language | English |
| last_indexed | 2024-03-12T00:07:43Z |
| publishDate | 2023-09-01 |
| publisher | Wiley |
| record_format | Article |
| series | CAAI Transactions on Intelligence Technology |
| spelling | doaj.art-feeb1f2a28144cef86de0f23547d9aa52023-09-16T16:19:35ZengWileyCAAI Transactions on Intelligence Technology2468-23222023-09-01831087110010.1049/cit2.12187A federated learning scheme meets dynamic differential privacyShengnan Guo0Xibin Wang1Shigong Long2Hai Liu3Liu Hai4Toong Hai Sam5State Key Laboratory of Public Big Data College of Computer Science and Technology Guizhou University Guiyang ChinaSchool of Big Data, Key Laboratory of Electric Power Big Data of Guizhou Province Guizhou Institute of Technology Guiyang ChinaState Key Laboratory of Public Big Data College of Computer Science and Technology Guizhou University Guiyang ChinaState Key Laboratory of Public Big Data College of Computer Science and Technology Guizhou University Guiyang ChinaSchool of Information Guizhou University of Finance and Economics Guiyang ChinaFaculty of Business and Communication INTI International University Nilai MalaysiaAbstract Federated learning is a widely used distributed learning approach in recent years, however, despite model training from collecting data become to gathering parameters, privacy violations may occur when publishing and sharing models. A dynamic approach is proposed to add Gaussian noise more effectively and apply differential privacy to federal deep learning. Concretely, it is abandoning the traditional way of equally distributing the privacy budget ϵ and adjusting the privacy budget to accommodate gradient descent federation learning dynamically, where the parameters depend on computation derived to avoid the impact on the algorithm that hyperparameters are created manually. It also incorporates adaptive threshold cropping to control the sensitivity, and finally, moments accountant is used to counting the ϵ consumed on the privacy‐preserving, and learning is stopped only if the ϵtotal by clients setting is reached, this allows the privacy budget to be adequately explored for model training. The experimental results on real datasets show that the method training has almost the same effect as the model learning of non‐privacy, which is significantly better than the differential privacy method used by TensorFlow.https://doi.org/10.1049/cit2.12187data privacymachine learningsecurity of data |
| spellingShingle | Shengnan Guo Xibin Wang Shigong Long Hai Liu Liu Hai Toong Hai Sam A federated learning scheme meets dynamic differential privacy CAAI Transactions on Intelligence Technology data privacy machine learning security of data |
| title | A federated learning scheme meets dynamic differential privacy |
| title_full | A federated learning scheme meets dynamic differential privacy |
| title_fullStr | A federated learning scheme meets dynamic differential privacy |
| title_full_unstemmed | A federated learning scheme meets dynamic differential privacy |
| title_short | A federated learning scheme meets dynamic differential privacy |
| title_sort | federated learning scheme meets dynamic differential privacy |
| topic | data privacy machine learning security of data |
| url | https://doi.org/10.1049/cit2.12187 |
| work_keys_str_mv | AT shengnanguo afederatedlearningschememeetsdynamicdifferentialprivacy AT xibinwang afederatedlearningschememeetsdynamicdifferentialprivacy AT shigonglong afederatedlearningschememeetsdynamicdifferentialprivacy AT hailiu afederatedlearningschememeetsdynamicdifferentialprivacy AT liuhai afederatedlearningschememeetsdynamicdifferentialprivacy AT toonghaisam afederatedlearningschememeetsdynamicdifferentialprivacy AT shengnanguo federatedlearningschememeetsdynamicdifferentialprivacy AT xibinwang federatedlearningschememeetsdynamicdifferentialprivacy AT shigonglong federatedlearningschememeetsdynamicdifferentialprivacy AT hailiu federatedlearningschememeetsdynamicdifferentialprivacy AT liuhai federatedlearningschememeetsdynamicdifferentialprivacy AT toonghaisam federatedlearningschememeetsdynamicdifferentialprivacy |