Skip to content

Secure outsourced matrix computation and application to neural networks

Authors: X. Jiang; M. Kim; K. E. Lauter; Y. Song
Year: 2018

type: technique collection: sok-references status: auto

Links: ePrint 2018/1041

Summary

Auto-extracted summary (unreviewed)

The paper presents a generic method to encrypt matrices homomorphically, with a new matrix encoding and efficient evaluation strategies for addition, multiplication and transposition, plus packing of several matrices into one ciphertext for better amortized cost. Built on this, the E2DM framework performs CNN prediction where both the data and the model are encrypted. With a CKKS implementation in Microsoft SEAL, multiplying two encrypted 64x64 matrices takes 0.6 s, transposition 0.09 s, and classifying 64 MNIST images takes 1.69 s (26 ms per image amortized).

Abstract

Homomorphic Encryption (HE) is a powerful cryptographic primitive to address privacy and security issues in outsourcing computation on sensitive data to an untrusted computation environment. Comparing to secure Multi-Party Computation (MPC), HE has advantages in supporting non-interactive operations and saving on communication costs. However, it has not come up with an optimal solution for modern learning frameworks, partially due to a lack of efficient matrix computation mechanisms. In this work, we present a practical solution to encrypt a matrix homomorphically and perform arithmetic operations on encrypted matrices. Our solution includes a novel matrix encoding method and an efficient evaluation strategy for basic matrix operations such as addition, multiplication, and transposition. We also explain how to encrypt more than one matrix in a single ciphertext, yielding better amortized performance. Our solution is generic in the sense that it can be applied to most of the existing HE schemes. It also achieves reasonable performance for practical use; for example, our implementation takes 0.6 seconds to multiply two encrypted square matrices of order 64 and 0.09 seconds to transpose a square matrix of order 64. Our secure matrix computation mechanism has a wide applicability to our new framework E2DM, which stands for encrypted data and encrypted model. To the best of our knowledge, this is the first work that supports secure evaluation of the prediction phase based on both encrypted data and encrypted model, whereas previous work only supported applying a plain model to encrypted data. As a benchmark, we report an experimental result to classify handwritten images using convolutional neural networks (CNN). Our implementation on the MNIST dataset takes 1.69 seconds to compute ten likelihoods of 64 input images simultaneously, yielding an amortized rate of 26 milliseconds per image.
Source: pdf

Extracted details

Field Value Source Confidence
contribution Novel matrix encoding and efficient homomorphic algorithms for matrix addition, multiplication and transposition llm 0.85
contribution Packing multiple matrices in a single ciphertext for better amortized performance llm 0.85
contribution E2DM framework: first secure CNN prediction on both encrypted data and encrypted model llm 0.85
contribution CKKS/SEAL implementation and MNIST benchmark with lower ciphertext size and latency than CryptoNets llm 0.85
schemes CKKS llm 0.85
libraries SEAL llm 0.85
models CNN llm 0.85
datasets MNIST llm 0.85
tasks secure outsourced matrix computation llm 0.85
tasks encrypted CNN inference (handwritten digit classification) llm 0.85
hardware MacBook Pro, Intel Core i9, 4 cores at 2.3 GHz llm 0.85
security level 80-bit llm 0.85
threat model untrusted cloud; encrypted data and encrypted model outsourced by data owner and model provider llm 0.85
interaction non-interactive FHE llm 0.85
code url https://github.com/keras-team/keras.21 (URL found in the full text) regex 0.50
code url https://github.com/Microsoft/SEAL (URL found in the full text) regex 0.50
llm relevance medium llm 0.85
key result encrypted matrix multiplication time: 0.6 s (two 64x64 encrypted matrices) llm 0.85
key result encrypted matrix transposition time: 0.09 s (64x64 matrix) llm 0.85
key result CNN inference latency: 1.69 s for 64 images (MNIST, encrypted data and encrypted model) llm 0.85
key result amortized inference time: 26 ms per image (MNIST, 64 images per ciphertext) llm 0.85
key result fresh ciphertext size: 260 KB (MatMult), 80 KB (MatAdd), 140 KB (MatTrans) (N=2^13, log q = 130/40/70) llm 0.85

SoK reference [79]

X. Jiang, M. Kim, K. E. Lauter, and Y. Song, “Secure outsourced matrix computation and application to neural networks,” in ACM CCS 2018: 25th Conference on Computer and Communications Security, D. Lie, M. Mannan, M. Backes, and X. Wang, Eds. Toronto, ON, Canada: ACM Press, Oct. 15–19, 2018, pp. 1209– 1222.

Listed on the reference page of the site.

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • scheme: MPC 10 [p.1], CKKS 2 [p.12]
  • library: Microsoft SEAL 3 [p.12], ABY / ABY3 1 [p.22]
  • model: CNN 12 [p.1], SVM 2 [p.2], Logistic regression 2 [p.22], AlexNet 1 [p.15]
  • dataset: MNIST 12 [p.1]
  • technique: Matrix multiplication 50 [p.2], Rotations 33 [p.2], SIMD packing 21 [p.2], Convolution 20 [p.1], Square activation 5 [p.3], RNS 5 [p.12], Key switching 3 [p.5], Rescaling 3 [p.12], Sigmoid 2 [p.15], Non-interactive 2 [p.1], Two-party computation 2 [p.21]
  • assumption: RLWE 4 [p.4], LWE 3 [p.4]
  • task: Language modeling / generation 5 [p.4], Medical / healthcare 4 [p.1], Genomics 4 [p.22]
  • topic: Homomorphic encryption 72 [p.1], Benchmark 3 [p.1], Open source 2 [p.21]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/sok-references/079_secure-outsourced-matrix-computation-and-application-to-neur.pdf sok-references 23 406.3 KB 184d8bfe3985

Database key: secure-outsourced-matrix-computation-and-application-to-neural-networks · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set secure-outsourced-matrix-computation-and-application-to-neural-networks COLUMN VALUE