Round-Efficient Protocols for Secure Multiparty Fixed-Point Arithmetic

Catrina, Octavian

doi:10.1109/iccomm.2018.8484794

Cited by 18 publications

(20 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Notice that the SGD algorithm is not over integers, while Shamir's secret sharing is built over finite fields like Z p . Recall that [2] and [3] presents secure fixed-point addition, multiplication, truncation and less-then-zero (LTZ) protocol with respect to Shamir's scheme. Thus we adopt [2] and [3] to realize the secure addition and multiplication, which details are recalled in Sect.…”

Section: Building Blocks Of Our Protocolmentioning

confidence: 99%

“…We now present a detailed overview of fixed-point calculations in [2] and [3], which also explains the notations to make it easier to read the following subsections. We will first introduce the representation of fixed-point numbers in Shamir's sharing and then recall the addition and multiplication with truncation, and finally introduce the less-than-zero protocol.…”

Section: Secure Fixed-point Calculation [2] and [3]mentioning

confidence: 99%

“…We will first introduce the representation of fixed-point numbers in Shamir's sharing and then recall the addition and multiplication with truncation, and finally introduce the less-than-zero protocol. More details of the protocols can be referred to [2] and [3] or Appendix B.…”

Section: Secure Fixed-point Calculation [2] and [3]mentioning

confidence: 99%

“…The Less-Than-Zero Protocol. In this subsection we introduce the LTZ protocol from [2] based on bit comparison and precise truncation Div2m for secure comparison. According to Formula 1, we need to decide whether a number is greater than 0.…”

Section: Secure Fixed-point Calculation [2] and [3]mentioning

confidence: 99%

“…We believe that the protocol will not leakage any additional information except normal output. Since our protocol is implemented by the secure computing framework from [3] and [2], the security of the protocol can also be reduced to their security. More specifically, we make the following claim and proof.…”

Section: Securitymentioning

confidence: 99%

See 4 more Smart Citations

Privacy-Preserving Support Vector Machines with Flexible Deployment and Error Correction

Huang

Ding

2021

Information Security Practice and Experience

View full text Add to dashboard Cite

Support vector machines (SVMs) are one of the most commonly used models for classification problems in machine learning. Nowadays there is an important scenario that many different parties jointly perform SVM training by integrating their individual data, while at the same time it is required that privacy of data can be preserved. At present there are three main routes to achieving privacy-preserving SVM. First, all parties jointly generate kernel matrices privately and then use them for remaining training (e.g. Yu et al. 2006). Second, based on the first route, an additional randomization is adopted to randomize kernel matrices in order to (heuristically) hide information exposed by kernel matrices (e.g. Mangasarian et al. 2008). Third, also the securest one, all parties run MPC protocols for computing whole optimization algorithms privately (not merely the generation of kernel matrices as the first two routes do) (e.g. Liu et al. 2018 andWang et al. 2020).In this paper we propose a new efficient privacy-preserving SVM protocol in the third route that privately realizes the gradient descent method to optimize SVM and its security is proven in the semi-honest model. Our protocol admits the following advantages.-The protocol is of flexible deployment. It supports the deployment of arbitrarily multiple servers and multiple clients. -The protocol can tolerate dropping-out of some servers.-The protocol admits the ability of malicious-error-message correction (which is actually beyond the semi-honest security). If a small number of messages are corrupted, it can still recover correct messages as desired. We remark that none of the above advantages can be obtained by some known work. Moreover, when compared to the privacy-preserving SVM by Liu et al. 2018 andWang et al. 2020, our protocol achieves higher efficiency. We implement our protocol in Python and the experiments verify its efficiency.

show abstract