Welcome to the resource topic for 2024/537
Title:
Confidential and Verifiable Machine Learning Delegations on the Cloud
Authors: Wenxuan Wu, Soamar Homsi, Yupeng Zhang
Abstract:With the growing adoption of cloud computing, the ability to store data and delegate computations to powerful and affordable cloud servers have become advantageous for both companies and individual users. However, the security of cloud computing has emerged as a significant concern. Particularly, Cloud Service Providers (CSPs) cannot assure data confidentiality and computations integrity in mission-critical applications. In this paper, we propose a confidential and verifiable delegation scheme that advances and overcomes major performance limitations of existing Secure Multiparty Computation (MPC) and Zero Knowledge Proof (ZKP). Secret-shared Data and delegated computations to multiple cloud servers remain completely confidential as long as there is at least one honest MPC server. Moreover, results are guaranteed to be valid even if all the participating servers are malicious. Specifically, we design an efficient protocol based on interactive proofs, such that most of the computations generating the proof can be done locally on each server. In addition, we propose a special protocol for matrix multiplication where the overhead of generating the proof is asymptotically smaller than the time to evaluate the result in MPC. Experimental evaluation demonstrates that our scheme significantly outperforms prior work, with the online prover time being 1-2 orders of magnitude faster. Notably, in the matrix multiplication protocol, only a minimal 2% of the total time is spent on the proof generation. Furthermore, we conducted tests on machine learning inference tasks. We executed the protocol for a fully-connected neural network with 3 layers on the MNIST dataset and it takes 2.6 seconds to compute the inference in MPC and generate the proof, 88× faster than prior work. We also tested the convolutional neural network of Lenet with 2 convolution layers and 3 dense layers and the running time is less than 300 seconds across three servers.
ePrint: https://eprint.iacr.org/2024/537
See all topics related to this paper.
Feel free to post resources that are related to this paper below.
Example resources include: implementations, explanation materials, talks, slides, links to previous discussions on other websites.
For more information, see the rules for Resource Topics .