Skip to content

Evaluation of Privacy-Preserving Support Vector Machine (SVM) Learning Using Homomorphic Encryption

Authors: William J. Buchanan; Hisham Ali
Year: 2025
Venue: Cryptography

type: benchmark collection: ppml status: auto

Links: DOI

Summary

Auto-extracted summary (unreviewed)

Journal version of Buchanan and Ali's study: an SVM model trained with scikit-learn is applied to homomorphically encrypted inputs with OpenFHE, and the overhead is measured while varying multiplication depth, scale size, first modulus size, security level, batch size and ring dimension for linear and polynomial kernels. Ring dimension and modulus size are the dominant cost factors: raising the ring dimension from 16,384 to 131,072 increases execution time from 915.2 s to 7716.4 s (8.4x) while accuracy stays constant, and kernel choice has little effect on HE overhead.

Abstract

The requirement for privacy-aware machine learning increases as we continue to use PII (personally identifiable information) within machine training. To overcome the existing privacy issues, we can apply fully homomorphic encryption (FHE) to encrypt data before they are fed into a machine learning model. This involves generating a homomorphic encryption key pair, where the public key encrypts the input data and the private key decrypts the output. However, there is often a performance hit when we use homomorphic encryption, so this paper evaluates the performance overhead of using an SVM (support vector machine) machine learning technique with the OpenFHE homomorphic encryption library. This uses Python and the scikit-learn library to create an SVM model, which can then be used with homomorphically encrypted data inputs and then produce a homomorphically encrypted result. The experiments include a range of variables, such as multiplication depth, scale size, first modulus size, security level, batch size, and ring dimension, along with two different SVM models, SVM-poly and SVM-linear. Overall, the results show that the two main parameters that affect performance are ring dimension and modulus size, and SVM-poly and SVM-linear show similar performance levels.
Source: crossref

Extracted details

Field Value Source Confidence
contribution privacy-preserving SVM implemented with FHE in OpenFHE llm 0.75
contribution empirical evaluation of feasibility and performance across encryption parameters and datasets llm 0.75
contribution analysis of the trade-off between computational efficiency and classification accuracy llm 0.75
schemes CKKS llm 0.75
libraries OpenFHE llm 0.75
libraries scikit-learn llm 0.75
models SVM (linear kernel) llm 0.75
models SVM (polynomial kernel) llm 0.75
tasks privacy-preserving SVM inference llm 0.75
tasks classification llm 0.75
security level 128-bit, 192-bit and 256-bit configurations evaluated llm 0.75
threat model outsourced computation on encrypted inputs; server does not hold the secret key llm 0.75
interaction non-interactive FHE llm 0.75
code url https://github.com/openfheorg/openfhe-genomic-examples/blob/main/demo-chi2.cpp (URL found in the full text) regex 0.50
code url https://github.com/openfheorg/openfhe-genomic-examples/blob/main/demo-logistic.cpp (URL found in the full text) regex 0.50
code url https://github.com/openfheorg/education/tree/main/examples/FHE_SVM_Examples (URL found in the full text) regex 0.50
code url https://github.com/openfheorg/education/tree/main/ llm 0.75
limitations homomorphic encryption introduces substantial computational overhead that must be managed by parameter tuning llm 0.75
llm relevance low llm 0.75
key result execution time: 915.2 s to 7716.4 s (8.4x increase) (ring dimension increased from 16,384 to 131,072) llm 0.75
key result accuracy: consistent across all parameter configurations (SVM-linear and SVM-poly on encrypted data) llm 0.75

Keyword mentions

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

  • scheme: CKKS 33 [p.2], BFV 10 [p.2], BGV 9 [p.2], TFHE 8 [p.2], FHEW 3 [p.2], MPC 2 [p.11], Partially homomorphic encryption 2 [p.1]
  • library: OpenFHE 13 [p.1], Microsoft SEAL 5 [p.3], PALISADE 1 [p.3], HElib 1 [p.23], Concrete (Zama) 1 [p.23], TensorFlow 1 [p.23]
  • model: SVM 120 [p.1], Linear regression 3 [p.7], Logistic regression 2 [p.2], Decision tree 1 [p.11]
  • dataset: Iris 13 [p.9], UCI repository 1 [p.9], Cora / Citeseer / PubMed (graphs) 1 [p.23]
  • hardware: GPU 5 [p.3], FPGA 3 [p.20], Intel Xeon 2 [p.10], CUDA 1 [p.3]
  • technique: Bootstrapping 16 [p.4], Chebyshev 6 [p.4], Rescaling 3 [p.5], SIMD packing 3 [p.20], Relinearization 2 [p.18]
  • assumption: Lattice-based 6 [p.2], RLWE 3 [p.6], LWE 3 [p.4]
  • task: Genomics 8 [p.6], Medical / healthcare 3 [p.1]
  • topic: Homomorphic encryption 84 [p.1], Private / secure inference 18 [p.9], Open source 8 [p.7], Privacy-preserving machine learning 6 [p.1], Benchmark 2 [p.11]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/ppml/inference/buchanan-2025_evaluation-of-privacy-preserving-svm-learning-using-he-journal.pdf ppml/inference 23 915.4 KB 76916d8de004 was Buchanan and Ali - 2025 - Evaluation of Privacy-Preserving Support Vector Machine (SVM) Learning Using Homomorphic Encryption.pdf

Database key: evaluation-of-privacy-preserving-support-vector-machine-svm-learning · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set evaluation-of-privacy-preserving-support-vector-machine-svm-learning COLUMN VALUE