Skip to content

On Fully Homomorphic Encryption for Privacy-Preserving Deep Learning

Year: 2019

type: application collection: ppml status: auto

Summary

Auto-extracted summary (unreviewed)

Implements CNN inference on FHE ciphertexts by expressing the network as logic circuits and measures the number of circuit operations required as a function of the numerical representation and of the security parameters. The authors find that full correct classification is reached with 6 fractional bits and characterise the trade-off between security level, prediction capability and computational cost.

Abstract

Given the rise of Machine Learning (ML) applications using sensitive private data, we present an implementation of Fully Homomorphic Encryption (FHE) with Convolutional Neural Networks (CNNs) for privacy-preserving Deep Learning (DL). This permits to utilize DL image recognition algorithms without compromising personal data of served customers, e.g medical images. The combination of FHE and CNN is accomplished by using logic circuits, which enable to conduct deep learning algorithms on ciphertext instead of plaintext. We provide a set of practical measurements from an implementation on the number of logic circuit operations required to carry out privacy-preserving ML. The results are function of the numerical representation and security parameters. Our results indicate that the protection of customer data is a trade-off between the required level of security, prediction capability and computational complexity. Thus, we reach full correct classification with 6 bits in the fractional part, and evaluate the time costs associated with our circuits as a function of its security parameters.
Source: pdf

Extracted details

Field Value Source Confidence
contribution FHE-based CNN inference via logic circuits llm 0.60
contribution measurements of logic-circuit operation counts versus numerical representation and security parameters llm 0.60
contribution trade-off analysis between security, prediction capability and computational complexity llm 0.60
schemes FHE (bit-wise logic circuits) llm 0.60
models CNN llm 0.60
tasks image classification llm 0.60
tasks privacy-preserving deep learning inference llm 0.60
threat model untrusted server computes on client's encrypted data llm 0.60
interaction non-interactive FHE llm 0.60
code url https://github.com/comsyslab/homomorphic-paper-code (URL found in the full text) regex 0.50
code url https://github.com/comsyslab/homomorphic-paper-code llm 0.60
limitations protection of data is a trade-off between security level, prediction capability and computational complexity llm 0.60
llm relevance low llm 0.60
key result fractional bits for full correct classification: 6 bits (CNN evaluated with FHE logic circuits) llm 0.60

Keyword mentions

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

  • scheme: BGV 3 [p.1], GSW 2 [p.1], Paillier 2 [p.1]
  • model: CNN 34 [p.1]
  • dataset: MNIST 5 [p.2]
  • technique: Convolution 15 [p.1], Bootstrapping 12 [p.1], Softmax 8 [p.3], ReLU 3 [p.3]
  • assumption: LWE 4 [p.1], Lattice-based 2 [p.1]
  • topic: Homomorphic encryption 19 [p.1], Benchmark 3 [p.2]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/surveys/marcano-2019_on-fhe-for-privacy-preserving-deep-learning.pdf ppml/surveys 6 184.6 KB eb7bb931d635 was Marcano et al. - 2019 - On Fully Homomorphic Encryption for Privacy-Preserving Deep Learning.pdf

Database key: on-fully-homomorphic-encryption-for-privacy-preserving-deep-learning · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set on-fully-homomorphic-encryption-for-privacy-preserving-deep-learning COLUMN VALUE