Fast Large-Scale Honest-Majority MPC for Malicious Adversaries

Chida, Koji; Genkin, Daniel; Hamada, Keisuke; Ikarashi, Dai; Kikuchi, Ryo; Lindell, Yehuda; Nof, Ariel

doi:10.1007/978-3-319-96878-0_2

Cited by 112 publications

(87 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Then we will prove that it holds for i = m. In Step 3, we calculate w m = x m − max m−1 . By correctness 18 To argue security, first observe that P 2 learns no information from the protocol (as F DReLU ({P 0 , P 1 }, P 2 ) and F SS ({P 0 , P 1 }, P 2 ) provides outputs only to P 0 and P 1 ). Now, P j , j ∈ {0, 1} only learn fresh shares of the values β i , max i , ind i and hence any information learned by either party can be perfectly simulated through appropriate shares of 0 (over Z L ).…”

Section: Divisionmentioning

confidence: 99%

“…Secure multi-party computation (MPC) [8,17,25,40] and specifically 3-party computation [6,9,10,14,18,23,28,32] provide a generic solution for the above problem. However, general purpose MPC/3PC work over low-level circuits (either arithmetic or boolean) and for complex tasks such as neural network training, this leads to highly inefficient protocols that can take "forever to execute" (also pointed out by [21,26,33,36]).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

SecureNN: 3-Party Secure Computation for Neural Network Training

Wagh

Gupta

Chandran

2019

Proceedings on Privacy Enhancing Technologies

293

320

View full text Add to dashboard Cite

Neural Networks (NN) provide a powerful method for machine learning training and inference. To effectively train, it is desirable for multiple parties to combine their data – however, doing so conflicts with data privacy. In this work, we provide novel three-party secure computation protocols for various NN building blocks such as matrix multiplication, convolutions, Rectified Linear Units, Maxpool, normalization and so on. This enables us to construct three-party secure protocols for training and inference of several NN architectures such that no single party learns any information about the data. Experimentally, we implement our system over Amazon EC2 servers in different settings. Our work advances the state-of-the-art of secure computation for neural networks in three ways: 1. Scalability: We are the first work to provide neural network training on Convolutional Neural Networks (CNNs) that have an accuracy of > 99% on the MNIST dataset; 2. Performance: For secure inference, our system outperforms prior 2 and 3-server works (SecureML, MiniONN, Chameleon, Gazelle) by 6×-113× (with larger gains obtained in more complex networks). Our total execution times are 2 − 4× faster than even just the online times of these works. For secure training, compared to the only prior work (SecureML) that considered a much smaller fully connected network, our protocols are 79× and 7× faster than their 2 and 3-server protocols. In the WAN setting, these improvements are more dramatic and we obtain an improvement of 553×! 3. Security: Our protocols provide two kinds of security: full security (privacy and correctness) against one semi-honest corruption and the notion of privacy against one malicious corruption [Araki et al. CCS’16]. All prior works only provide semi-honest security and ours is the first system to provide any security against malicious adversaries for the secure computation of complex algorithms such as neural network inference and training. Our gains come from a significant improvement in communication through the elimination of expensive garbled circuits and oblivious transfer protocols.

show abstract

Section: Divisionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

SecureNN: 3-Party Secure Computation for Neural Network Training

Wagh

Gupta

Chandran

2019

Proceedings on Privacy Enhancing Technologies

293

320

View full text Add to dashboard Cite

show abstract

“…However, options for non-continuous crossing alongside options for continuous crossing are offered by some venues, such as LSE's Turquoise and ITG's Posit. Turquoise offers the option of randomised period crossing (where matching is at random intervals between 10-45 seconds, depending upon liquidity) 6 , while the option Posit Match matches orders during a 30-second interval several times a day at pre-determined times [20, p.39].…”

Section: Trading In the Darkmentioning

confidence: 99%

“…Special dark order types that enable volume to be hidden on a lit exchange, e.g., an Iceberg Order, are not included. 6 Turqoise Plato Uncross: https://www.lseg.com/sites/default/files/content/documents/ TURQUOISE_UNCROSS_FACTSHEET_2016_AW.pdf Interestingly, markets have quickly adapted to the DVC by moving volume away from dark pools to newly-popular periodic auction venues [10]. Unlike continuous central limit order books (CLOBs), where trades are executed instantly as soon as there is a match, in a periodic auction, matching only occurs at the end of a call period.…”

Section: Out Of the Dark And Into The Semi-transparentmentioning

confidence: 99%

“…The other secret sharing scheme we consider is that of Shamir sharing with three players. Here the SCALE system implements the methodoloy from [14], this latter methodology results in a faster online phase then for example using the methodology in [6] which provides a shorter overall run time (at the expense of more expensive online computation).…”

Section: Mpc Backgroundmentioning

confidence: 99%

See 1 more Smart Citation

MPC Joins The Dark Side

Cartlidge

Smart

Alaoui

2019

Proceedings of the 2019 ACM Asia Conference on Computer and Communications Security

View full text Add to dashboard Cite

We consider the issue of securing dark pools/markets in the financial services sector. These markets currently are executed via trusted third parties, leading to potential fraud being able to be conducted by the market operators. We present a potential solution to this problem by using Multi-Party Computation to enable a trusted third party to be emulated in software. Our experiments show that whilst the standard market clearing mechanism of Continuous Double Auction in lit markets is not currently viable when executed using MPC, a popular mechanism for evaluating dark markets, namely the volume matching methodology, is viable. We present experimental validation of this conclusion by presenting the expected throughputs for such markets in two popular MPC paradigms; namely the two party dishonest majority setting and the honest majority three party setting.

show abstract

Secure Multiparty Computation

Lindell¹

2022

Asymmetric Cryptography

View full text Add to dashboard Cite

Fast Large-Scale Honest-Majority MPC for Malicious Adversaries

Cited by 112 publications

References 25 publications

SecureNN: 3-Party Secure Computation for Neural Network Training

SecureNN: 3-Party Secure Computation for Neural Network Training

MPC Joins The Dark Side

Secure Multiparty Computation

Contact Info

Product

Resources

About