SPEED: secure, private, and efficient deep learning
From MaRDI portal
Publication:2051270
Abstract: We introduce a deep learning framework able to deal with strong privacy constraints. Based on collaborative learning, differential privacy and homomorphic encryption, the proposed approach advances state-of-the-art of private deep learning against a wider range of threats, in particular the honest-but-curious server assumption. We address threats from both the aggregation server, the global model and potentially colluding data holders. Building upon distributed differential privacy and a homomorphic argmax operator, our method is specifically designed to maintain low communication loads and efficiency. The proposed method is supported by carefully crafted theoretical results. We provide differential privacy guarantees from the point of view of any entity having access to the final model, including colluding data holders, as a function of the ratio of data holders who kept their noise secret. This makes our method practical to real-life scenarios where data holders do not trust any third party to process their datasets nor the other data holders. Crucially the computational burden of the approach is maintained reasonable, and, to the best of our knowledge, our framework is the first one to be efficient enough to investigate deep learning applications while addressing such a large scope of threats. To assess the practical usability of our framework, experiments have been carried out on image datasets in a classification context. We present numerical results that show that the learning procedure is both accurate and private.
Recommendations
- Privacy preserving multi-party computation delegation for deep learning in cloud computing
- Privacy-preserving distributed deep learning based on secret sharing
- EPIC: efficient private image classification (or: learning from the masters)
- Fast homomorphic evaluation of deep discretized neural networks
- Maliciously secure matrix multiplication with applications to private deep learning
Cites work
- (Leveled) fully homomorphic encryption without bootstrapping
- Advances in Cryptology - CRYPTO 2003
- Extremal mechanisms for local differential privacy
- Faster fully homomorphic encryption: bootstrapping in less than 0.1 seconds
- Faster homomorphic encryption is not enough: improved heuristic for multiplicative depth minimization of Boolean circuits
- scientific article; zbMATH DE number 1614382 (Why is no real title available?)
- ML confidential: machine learning on encrypted data
- Our Data, Ourselves: Privacy Via Distributed Noise Generation
- Privacy-Preserving Stream Aggregation with Fault Tolerance
- The algorithmic foundations of differential privacy
- What can we learn privately?
Cited in
(9)- Privacy preserving multi-party computation delegation for deep learning in cloud computing
- Fast secure comparison for medium-sized integers and its application in binarized neural networks
- Privacy-preserving distributed deep learning based on secret sharing
- Maliciously secure matrix multiplication with applications to private deep learning
- Secure transfer learning for machine fault diagnosis under different operating conditions
- \textsf{ComBo}: a novel functional bootstrapping method for efficient evaluation of nonlinear functions in the encrypted domain
- PD-ML-Lite: Private Distributed Machine Learning from Lightweight Cryptography
- FLSwitch: towards secure and fast model aggregation for federated deep learning with a learning state-aware switch
- Relations among new CCA security notions for approximate FHE
This page was built for publication: SPEED: secure, private, and efficient deep learning
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2051270)
