Skip to content

Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities

Authors: Bernardo Pulido-Gaytan; Andrei Tchernykh; Jorge M. Cortés-Mendoza; Mikhail Babenko; Gleb Radchenko; Arutyun Avetisyan; Alexander Yu Drozdov
Year: 2021
Venue: Peer-to-Peer Networking and Applications

type: survey collection: ppml status: auto

Links: DOI

Summary

Auto-extracted summary (unreviewed)

A survey of homomorphic encryption for privacy-preserving neural networks in the cloud, covering FHE fundamentals (bootstrapping, key switching, noise, multiplicative depth), algorithmic and running errors that arise when encoding real numbers and approximating activations, available libraries and frameworks (SEAL, HElib, TFHE, PALISADE, cuHE, HEAAN, HE-transformer), application domains, and open challenges. It also compares three NN-HE implementations against CryptoNets and LoLa using high-level and low-level tools.

Abstract

Classical machine learning modeling demands considerable computing power for internal calculations and training with big data in a reasonable amount of time. In recent years, clouds provide services to facilitate this process, but it introduces new security threats of data breaches. Modern encryption techniques ensure security and are considered as the best option to protect stored data and data in transit from an unauthorized third-party. However, a decryption process is necessary when the data must be processed or analyzed, falling into the initial problem of data vulnerability. Fully Homomorphic Encryption (FHE) is considered the holy grail of cryptography. It allows a non-trustworthy third-party resource to process encrypted information without disclosing confidential data. In this paper, we analyze the fundamental concepts of FHE, practical implementations, state-of-the-art approaches, limitations, advantages, disadvantages, potential applications, and development tools focusing on neural networks. In recent years, FHE development demonstrates remarkable progress. However, current literature in the homomorphic neural networks is almost exclusively addressed by practitioners looking for suitable implementations. It still lacks comprehensive and more thorough reviews. We focus on the privacy-preserving homomorphic encryption cryptosystems targeted at neural networks identifying current solutions, open issues, challenges, opportunities, and potential research directions.
Source: crossref

Extracted details

Field Value Source Confidence
contribution Review of HE cryptosystems at the intersection of cryptography and neural networks llm 0.80
contribution Accessible treatment of bootstrapping, key switching, noise, algorithmic and running errors llm 0.80
contribution Overview of HE libraries, NN frameworks and application domains llm 0.80
contribution Comparison of three NN-HE implementations with CryptoNets and LoLa llm 0.80
contribution Open problems and research directions llm 0.80
schemes CKKS llm 0.80
schemes BFV llm 0.80
schemes BGV llm 0.80
schemes TFHE llm 0.80
libraries SEAL llm 0.80
libraries HElib llm 0.80
libraries TFHE llm 0.80
libraries PALISADE llm 0.80
libraries cuHE llm 0.80
libraries HEAAN llm 0.80
libraries HE-transformer llm 0.80
tasks private neural network inference llm 0.80
tasks cloud machine learning llm 0.80
threat model non-trustworthy cloud third party processing encrypted data llm 0.80
interaction non-interactive FHE llm 0.80
code url https://github.com/karpathy/convnetjs (URL found in the full text) regex 0.50
code url https://github.com/amzn/amazon-dsstne (URL found in the full text) regex 0.50
code url https://github.com/NervanaSystems/neon (URL found in the full text) regex 0.50
llm relevance low llm 0.80

Keyword mentions

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

  • scheme: Leveled / somewhat HE 17 [p.3], Partially homomorphic encryption 12 [p.3], CKKS 11 [p.5], BFV 6 [p.5], TFHE 5 [p.15], NTRU 5 [p.5], BGN 5 [p.3], BGV 4 [p.5], Paillier 3 [p.3], RSA 3 [p.3], MPC 3 [p.15], GSW 2 [p.15]
  • library: Microsoft SEAL 16 [p.15], PyTorch 12 [p.19], PALISADE 9 [p.14], cuHE 8 [p.15], HElib 6 [p.15], TensorFlow 3 [p.15], CrypTen 2 [p.19], TenSEAL 1 [p.19], nGraph-HE 1 [p.25]
  • model: CNN 12 [p.8], AlexNet 3 [p.8], Logistic regression 3 [p.22], Decision tree 1 [p.22], Naive Bayes 1 [p.25]
  • dataset: MNIST 14 [p.8]
  • hardware: GPU 14 [p.8], FPGA 5 [p.11], ASIC 5 [p.11], CUDA 4 [p.15]
  • technique: Bootstrapping 51 [p.2], Key switching 11 [p.2], Convolution 11 [p.8], Polynomial approximation 6 [p.8], ReLU 6 [p.8], RNS 6 [p.9], Sigmoid 4 [p.8], SIMD packing 4 [p.13], Non-interactive 4 [p.8], Compiler 3 [p.15], NTT 2 [p.15], Encrypted training 2 [p.11]
  • assumption: Lattice-based 11 [p.4], LWE 5 [p.5], RLWE 4 [p.5], Post-quantum 3 [p.2]
  • task: Medical / healthcare 10 [p.2], Finance 10 [p.7], Genomics 7 [p.7]
  • topic: Homomorphic encryption 252 [p.1], Open source 4 [p.15]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/surveys/pulido-gaytan-2021_privacy-preserving-neural-networks-with-he-challenges-and-opportunities.pdf ppml/surveys 26 1.1 MB 2219e7710528 was Pulido-Gaytán et al. - 2021 - Privacy-preserving neural networks with Homomorphic encryption Challenges and opportunities.pdf

Database key: privacy-preserving-neural-networks-with-homomorphic-encryption · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set privacy-preserving-neural-networks-with-homomorphic-encryption COLUMN VALUE