Skip to content

SoK: Privacy-preserving computation techniques for deep learning

Authors: José Cabrero-Holgueras; Sergio Pastrana
Year: 2021
Venue: Proceedings on Privacy Enhancing Technologies

type: survey collection: sok-references status: auto

Links: DOI

Summary

Auto-extracted summary (unreviewed)

This SoK reviews how homomorphic encryption and secure multi-party computation have been adapted to deep learning training and inference, building a taxonomy of privacy requirements, techniques and architectures and then analyzing the state of the art by privacy goal, architecture, efficiency and usability. It highlights efficiency shortcomings, reproducibility problems from the lack of standard tools and interfaces, and poor integration with mainstream DL frameworks, to explain the gap between research proposals and practical MLaaS deployment.

Abstract

Deep Learning (DL) is a powerful solution for complex problems in many disciplines such as finance, medical research, or social sciences. Due to the high computational cost of DL algorithms, data scientists often rely upon Machine Learning as a Service (MLaaS) to outsource the computation onto third-party servers. However, outsourcing the computation raises privacy concerns when dealing with sensitive information, e.g., health or financial records. Also, privacy regulations like the European GDPR limit the collection, distribution, and use of such sensitive data. Recent advances in privacy-preserving computation techniques (i.e., Homomorphic Encryption and Secure Multiparty Computation) have enabled DL training and inference over protected data. However, these techniques are still immature and difficult to deploy in practical scenarios. In this work, we review the evolution of the adaptation of privacy-preserving computation techniques onto DL, to understand the gap between research proposals and practical applications. We highlight the relative advantages and disadvantages, considering aspects such as efficiency shortcomings, reproducibility issues due to the lack of standard tools and programming interfaces, or lack of integration with DL frameworks commonly used by the data science community.
Source: crossref

Extracted details

Field Value Source Confidence
contribution High-level taxonomy of privacy requirements, techniques and architectures for privacy-preserving deep learning llm 0.80
contribution Review of the evolution of HE- and SMPC-based deep learning proposals llm 0.80
contribution Analysis of proposals by privacy goal, architecture, efficiency and usability, including open-source availability and code maintenance llm 0.80
contribution Identification of recurring flaws and the gap between research and practical deployment llm 0.80
schemes homomorphic encryption llm 0.80
schemes secure multi-party computation llm 0.80
libraries SEAL llm 0.80
libraries HElib llm 0.80
libraries TF Encrypted llm 0.80
tasks privacy-preserving deep learning training and inference llm 0.80
tasks MLaaS llm 0.80
code url https://github.com/fchollet/keras (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 0.80

SoK reference [24]

J. Cabrero-Holgueras and S. Pastrana, “SoK: Privacy-preserving computation techniques for deep learning,” Proceedings on Privacy Enhancing Technologies, vol. 2021, no. 4, pp. 139–162, Oct. 2021.

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: Secret sharing 18 [p.7], MPC 16 [p.2], TEE / confidential computing 8 [p.4], Oblivious transfer 6 [p.7], CKKS 5 [p.6], TFHE 5 [p.6], Garbled circuits 5 [p.7], Zero-knowledge proofs 5 [p.7], Leveled / somewhat HE 4 [p.6], Paillier 3 [p.6], Partially homomorphic encryption 3 [p.6], BFV 2 [p.6], ElGamal 2 [p.6], Multi-key HE 2 [p.13], Scheme switching 2 [p.8]
  • library: ABY / ABY3 15 [p.8], EVA / CHET 9 [p.11], EzPC / CrypTFlow 9 [p.8], MP-SPDZ 6 [p.7], PyTorch 4 [p.11], Microsoft SEAL 3 [p.6], nGraph-HE 2 [p.21], HElib 1 [p.6], TensorFlow 1 [p.11]
  • model: CNN 4 [p.9], Linear regression 3 [p.9], Logistic regression 2 [p.9], GAN 1 [p.18]
  • dataset: MNIST 1 [p.10]
  • hardware: GPU 1 [p.4]
  • technique: Compiler 14 [p.8], Semi-honest model 10 [p.3], Bootstrapping 8 [p.6], ReLU 8 [p.9], Two-party computation 7 [p.8], Malicious model 7 [p.7], Federated learning 6 [p.4], Quantization 4 [p.9], Sigmoid 3 [p.9], Relinearization 2 [p.6], SIMD packing 2 [p.11], Convolution 2 [p.22], Pruning 2 [p.9], Model inversion 2 [p.15]
  • assumption: LWE 3 [p.6], Lattice-based 3 [p.8]
  • task: Medical / healthcare 6 [p.1]
  • topic: Homomorphic encryption 70 [p.2], Open source 41 [p.5], Private / secure inference 4 [p.11], Privacy-preserving machine learning 3 [p.18], Benchmark 3 [p.16]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/sok-references/024_sok-privacy-preserving-computation-techniques-for-deep-learn.pdf sok-references 24 539.6 KB da466e26a502

Database key: sok-privacy-preserving-computation-techniques-for-deep-learning · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set sok-privacy-preserving-computation-techniques-for-deep-learning COLUMN VALUE