A Secure Federated Learning framework using Homomorphic Encryption and Verifiable Computing

Madi, Abbass; Stan, Oana; Mayoue, Aurélien; Grivet-Sebert, Arnaud; Gouy-Pailler, Cédric; Sirdey, Renaud

doi:10.1109/rdaaps48126.2021.9452005

Cited by 33 publications

(21 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…FEMNIST, the federated version, was built by partitioning the data based on the writer [8]. The network architecture is the same as in [30]: a standard CNN composed of two convolution layers (respectively with 5 * 5 kernel size and 128 channels, and with 3 * 3 kernel size and 64 channels, each followed with 2 * 2 max pooling), a fully connected layer with 128 units and ReLu activation, and a final softmax output layer (486,654 parameters).…”

Section: Resultsmentioning

confidence: 99%

“…Table 1 shows the influence on the model accuracy of the adaptations necessary to ensure DP. Starting from a non-DP baseline from the state of the art [30], we successively modified parameters of the framework, each of these modifications being required by the DP analysis 7 . The successive steps are:…”

Section: Resultsmentioning

confidence: 99%

“…In terms of concrete HE, the fact that we are considering the simple Feder-atedAveraging operator allows us to use additive-only schemes such as in [30] where the Paillier cryptosystem is used with batching. In the experimental results reported in the next section we have used the BFV cryptosystem which allows for more massive batching and, as such, results in much lower (amortised) overheads.…”

Section: Homomorphic Encryption Protects the Data (And The Model) Aga...mentioning

confidence: 99%

“…In the experimental results reported in the next section we have used the BFV cryptosystem which allows for more massive batching and, as such, results in much lower (amortised) overheads. Additionally, one key contribution of [30] was to associate Paillierbased homomorphic calculations to Verifiable Computing (VC) techniques (e.g. [14]) to further extend the server threat model beyond the honest-but-curious one and bring execution integrity, as [29] did with BFV scheme.…”

Section: Homomorphic Encryption Protects the Data (And The Model) Aga...mentioning

confidence: 99%

“…Specifically, homomorphic encryption (HE) [37,45] or multi-party computation [6,36] allow to perform secure, blind computations without access to the inputs, and, in the case of HE, the outputs either. Additionally, HE can in some cases be associated to tools such as verifiable computing to bring additional computation integrity guarantees, including in the context of federated learning [30].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Protecting Data from all Parties: Combining FHE and DP in Federated Learning

Sébert¹,

Sirdey²,

Stan³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

This paper tackles the problem of ensuring training data privacy in a federated learning context. Relying on Homomorphic Encryption (HE) and Differential Privacy (DP), we propose a framework addressing threats on the privacy of the training data. Notably, the proposed framework ensures the privacy of the training data from all actors of the learning process, namely the data owners and the aggregating server. More precisely, while HE blinds a semi-honest server during the learning protocol, DP protects the data from semi-honest clients participating in the training process as well as end-users with black-box or white-box access to the trained model. In order to achieve this, we provide new theoretical and practical results to allow these techniques to be rigorously combined. In particular, by means of a novel stochastic quantisation operator, we prove DP guarantees in a context where the noise is quantised and bounded due to the use of HE. The paper is concluded by experiments which show the practicality of the entire framework in terms of both model quality (impacted by DP) and computational overhead (impacted by HE).

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Homomorphic Encryption Protects the Data (And The Model) Aga...mentioning

confidence: 99%