FLVoogd: Robust And Privacy Preserving Federated Learning

Tian Yuhang, Wang Rui, Qiao Yanqi, Panaousis Emmanouil, Liang Kaitai
Proceedings of The 14th Asian Conference on Machine Learning, PMLR 189:1022-1037, 2023.

Abstract

In this work, we propose FLVoogd, an updated federated learning method in which servers and clients collaboratively eliminate Byzantine attacks while preserving privacy. In particular, servers use automatic Density-based Spatial Clustering of Applications with Noise (DBSCAN) combined with Secure Multi-party Computation (SMPC) to cluster the benign majority without acquiring sensitive personal information. Meanwhile, clients build dual models and perform test-based distance controlling to adjust their local models toward the global one to achieve personalizing. Our framework is automatic and adaptive that servers/clients don’t need to tune the parameters during the training. In addition, our framework leverages SMPC’s operations, including multiplications, additions, and comparisons, where costly operations, like division and square root, are not required. Evaluations are carried out on some conventional datasets from the image classification field. The result shows that FLVoogd can effectively reject malicious uploads in most scenarios; meanwhile, it avoids data leakage from the server side.

Cite this Paper

BibTeX
@InProceedings{pmlr-v189-yuhang23a, title = {FLVoogd: Robust And Privacy Preserving Federated Learning}, author = {Yuhang, Tian and Rui, Wang and Yanqi, Qiao and Emmanouil, Panaousis and Kaitai, Liang}, booktitle = {Proceedings of The 14th Asian Conference on Machine Learning}, pages = {1022--1037}, year = {2023}, editor = {Khan, Emtiyaz and Gonen, Mehmet}, volume = {189}, series = {Proceedings of Machine Learning Research}, month = {12--14 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v189/yuhang23a/yuhang23a.pdf}, url = {https://proceedings.mlr.press/v189/yuhang23a.html}, abstract = {In this work, we propose FLVoogd, an updated federated learning method in which servers and clients collaboratively eliminate Byzantine attacks while preserving privacy. In particular, servers use automatic Density-based Spatial Clustering of Applications with Noise (DBSCAN) combined with Secure Multi-party Computation (SMPC) to cluster the benign majority without acquiring sensitive personal information. Meanwhile, clients build dual models and perform test-based distance controlling to adjust their local models toward the global one to achieve personalizing. Our framework is automatic and adaptive that servers/clients don’t need to tune the parameters during the training. In addition, our framework leverages SMPC’s operations, including multiplications, additions, and comparisons, where costly operations, like division and square root, are not required. Evaluations are carried out on some conventional datasets from the image classification field. The result shows that FLVoogd can effectively reject malicious uploads in most scenarios; meanwhile, it avoids data leakage from the server side.} }
Endnote
%0 Conference Paper %T FLVoogd: Robust And Privacy Preserving Federated Learning %A Tian Yuhang %A Wang Rui %A Qiao Yanqi %A Panaousis Emmanouil %A Liang Kaitai %B Proceedings of The 14th Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Emtiyaz Khan %E Mehmet Gonen %F pmlr-v189-yuhang23a %I PMLR %P 1022--1037 %U https://proceedings.mlr.press/v189/yuhang23a.html %V 189 %X In this work, we propose FLVoogd, an updated federated learning method in which servers and clients collaboratively eliminate Byzantine attacks while preserving privacy. In particular, servers use automatic Density-based Spatial Clustering of Applications with Noise (DBSCAN) combined with Secure Multi-party Computation (SMPC) to cluster the benign majority without acquiring sensitive personal information. Meanwhile, clients build dual models and perform test-based distance controlling to adjust their local models toward the global one to achieve personalizing. Our framework is automatic and adaptive that servers/clients don’t need to tune the parameters during the training. In addition, our framework leverages SMPC’s operations, including multiplications, additions, and comparisons, where costly operations, like division and square root, are not required. Evaluations are carried out on some conventional datasets from the image classification field. The result shows that FLVoogd can effectively reject malicious uploads in most scenarios; meanwhile, it avoids data leakage from the server side.
APA
Yuhang, T., Rui, W., Yanqi, Q., Emmanouil, P. & Kaitai, L.. (2023). FLVoogd: Robust And Privacy Preserving Federated Learning. Proceedings of The 14th Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 189:1022-1037 Available from https://proceedings.mlr.press/v189/yuhang23a.html.

Related Material