| Presentation | 2022-11-24 Social Surplus Maximization Using Incentive Mechanism for Cross-Silo Federated Learning with Differential Privacy Shota Miyagoshi, Takuji Tachibana, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In cross-silo federated learning, where multiple organizations participate, the prediction accuracy of the global model is improved by increasing the training data when a larger number of organizations participate in the learning of the local model. Each organization protects the privacy of its own data by using the training data only internally and disclosing and sharing the machine learning parameters. However, there is a possibility that the training data may be reconstructed from the machine learning parameters. This reconstruction can be avoided by using differential privacy. On the other hand, the prediction accuracy of the global model is reduced by using the differential privacy. The differential privacy of other organizations may reduce the number of participating organizations by decreasing the obtained revenue. Therefore, in this paper, we propose an incentive mechanism to maximize social surplus in cross-silo federated learning with differential privacy. The mechanism encourages active participation in cross-silo federated learning by transferring money between organizations. We evaluate the performance of the proposed method by simulation and show that each organization cooperates to maximize the social surplus while protecting privacy. We also examine whether partial participation of each organization in federated learning can further increase social surplus. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Federated learning / Differential privacy / Incentive mechanism / Distributed algorithm |
| Paper # | NS2022-101 |
| Date of Issue | 2022-11-17 (NS) |
| Conference Information | |
| Committee | NS / ICM / CQ |
|---|---|
| Conference Date | 2022/11/24(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Humanities and Social Sciences Center, Fukuoka Univ. + Online |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Network quality, Network measurement/management, Network virtualization, Network service, Blockchain, Security, Network intelligence/AI, etc. |
| Chair | Tetsuya Oishi(NTT) / Yuji Nomura(Fujitsu) / Jun Okamoto(NTT) |
| Vice Chair | Takumi Miyoshi(Shibaura Insti of Tech.) / Yu Miyoshi(NTT) / Eiji Takahashi(NEC) / Takefumi Hiraguri(Nippon Inst. of Tech.) / Gou Hasegawa(Tohoku Univ.) |
| Secretary | Takumi Miyoshi(NTT) / Yu Miyoshi(Kogakuin Univ.) / Eiji Takahashi(NTT) / Takefumi Hiraguri(Fujitsu) / Gou Hasegawa(NTT) |
| Assistant | Kotaro Mihara(NTT) / Ryo Yamamoto(Univ. of Electro-Comm) / Kimiko Kawashima(NTT) / Ryo Nakamura(Fukuoka Univ.) / Toshiro Nakahira(NTT) / Kenta Tsukatsune(Tokyo Metroplitan Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Network Systems / Technical Committee on Information and Communication Management / Technical Committee on Communication Quality |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Social Surplus Maximization Using Incentive Mechanism for Cross-Silo Federated Learning with Differential Privacy |
| Sub Title (in English) | |
| Keyword(1) | Federated learning |
| Keyword(2) | Differential privacy |
| Keyword(3) | Incentive mechanism |
| Keyword(4) | Distributed algorithm |
| 1st Author's Name | Shota Miyagoshi |
| 1st Author's Affiliation | University of Fukui(Univ. Fukui) |
| 2nd Author's Name | Takuji Tachibana |
| 2nd Author's Affiliation | University of Fukui(Univ. Fukui) |
| Date | 2022-11-24 |
| Paper # | NS2022-101 |
| Volume (vol) | vol.122 |
| Number (no) | NS-274 |
| Page | pp.pp.7-12(NS), |
| #Pages | 6 |
| Date of Issue | 2022-11-17 (NS) |