Secure Quantized Training for Deep Learning

Files

keller22a.pdf (484.66 KB)
  (Published version)

Date

2022

Authors

Keller, M.
Sun, K.

Editors

K, K.
Jegelka, S.
Song, L.
Szepesvari, C.
Niu, G.
Sabato, S.

Advisors

Journal Title

Journal ISSN

Volume Title

Type:

Conference paper

Citation

Proceedings of Machine Learning Research, 2022 / K, K., Jegelka, S., Song, L., Szepesvari, C., Niu, G., Sabato, S. (ed./s), pp.10912-10938

Statement of Responsibility

Conference Name

39th International Conference on Machine Learning, ICML 2022 (17 Jul 2022 - 23 Jul 2022 : Baltimore, Maryland, USA, PMLR 162)

Abstract

We implement training of neural networks in secure multi-party computation (MPC) using quantization commonly used in said setting. We are the first to present an MNIST classifier purely trained in MPC that comes within 0.2 percent of the accuracy of the same convolutional neural network trained via plaintext computation. More concretely, we have trained a network with two convolutional and two dense layers to 99.2% accuracy in 3.5 hours (under one hour for 99% accuracy). We have also implemented AlexNet for CIFAR-10, which converges in a few hours. We develop novel protocols for exponentiation and inverse square root. Finally, we present experiments in a range of MPC security models for up to ten parties, both with honest and dishonest majority as well as semi-honest and malicious security.

School/Discipline

Dissertation Note

Provenance

Description

Access Status

Rights

Copyright 2022 by the author(s)

License

Grant ID

Call number

Persistent link to this record