Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Secure Inference for Vertically Partitioned Data Using Multiparty Homomorphic Encryption

Shuangyi Chen, Yue Ju, Zhongwen Zhu, Ashish Khisti

TL;DR

The paper tackles secure inference when data are vertically partitioned across multiple clients and the model resides on a server. It introduces an end-to-end protocol based on Multiparty Homomorphic Encryption (MPHE) and a packing scheme called V-Pack to assemble a full encrypted input from per-client parts without transmitting encrypted model weights. Building on POSEIDON, it implements a two-phase workflow with a target key pair to enable end-to-end decryption by a coordinator, preserving data and model confidentiality even under collusion of up to $N-1$ parties. Experimental results on a CNN with MNIST show the approach is feasible, with setup costs dominating but inference overhead modest and scalable linearly with the number of clients.

Abstract

We propose a secure inference protocol for a distributed setting involving a single server node and multiple client nodes. We assume that the observed data vector is partitioned across multiple client nodes while the deep learning model is located at the server node. Each client node is required to encrypt its portion of the data vector and transmit the resulting ciphertext to the server node. The server node is required to collect the ciphertexts and perform inference in the encrypted domain. We demonstrate an application of multi-party homomorphic encryption (MPHE) to satisfy these requirements. We propose a packing scheme, that enables the server to form the ciphertext of the complete data by aggregating the ciphertext of data subsets encrypted using MPHE. While our proposed protocol builds upon prior horizontal federated training protocol~\cite{sav2020poseidon}, we focus on the inference for vertically partitioned data and avoid the transmission of (encrypted) model weights from the server node to the client nodes.

Secure Inference for Vertically Partitioned Data Using Multiparty Homomorphic Encryption

TL;DR

The paper tackles secure inference when data are vertically partitioned across multiple clients and the model resides on a server. It introduces an end-to-end protocol based on Multiparty Homomorphic Encryption (MPHE) and a packing scheme called V-Pack to assemble a full encrypted input from per-client parts without transmitting encrypted model weights. Building on POSEIDON, it implements a two-phase workflow with a target key pair to enable end-to-end decryption by a coordinator, preserving data and model confidentiality even under collusion of up to parties. Experimental results on a CNN with MNIST show the approach is feasible, with setup costs dominating but inference overhead modest and scalable linearly with the number of clients.

Abstract

We propose a secure inference protocol for a distributed setting involving a single server node and multiple client nodes. We assume that the observed data vector is partitioned across multiple client nodes while the deep learning model is located at the server node. Each client node is required to encrypt its portion of the data vector and transmit the resulting ciphertext to the server node. The server node is required to collect the ciphertexts and perform inference in the encrypted domain. We demonstrate an application of multi-party homomorphic encryption (MPHE) to satisfy these requirements. We propose a packing scheme, that enables the server to form the ciphertext of the complete data by aggregating the ciphertext of data subsets encrypted using MPHE. While our proposed protocol builds upon prior horizontal federated training protocol~\cite{sav2020poseidon}, we focus on the inference for vertically partitioned data and avoid the transmission of (encrypted) model weights from the server node to the client nodes.
Paper Structure (23 sections, 2 theorems, 4 figures, 1 table)

This paper contains 23 sections, 2 theorems, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Multiparty Homomorphic Encryption
  4. Secure Aggregation Using MPHE
  5. Setup Phase
  6. Aggregation Phase
  7. Decryption Phase
  8. Prior Work: POSEIDON
  9. System Overview
  10. System Setup
  11. Privacy Goals and Threat Model
  12. The Protocol
  13. Setup phase
  14. Inference Phase
  15. Secure Aggregation for Input Concatenation
  16. ...and 8 more sections

Key Result

Proposition 4.1

One of the output of the setup phase, $cpk$, is a public key corresponding to secret key $csk = \sum_i sk_i$.

Figures (4)

  • Figure 1: System setup
  • Figure 2: POSEIDON's structure
  • Figure 3: Packing scheme for vertical partitioned input.
  • Figure 4: Time and communication cost of some steps vs. The number of clients

Theorems & Definitions (2)

  • Proposition 4.1
  • Theorem 4.1