Skip to content

Privft: Private and fast text classification with homomorphic encryption

Authors: Ahmad Al Badawi; Luong Hoang; Chan Fook Mun; Kim Laine; Khin Mi Mi Aung
Year: 2020
Venue: IEEE Access, 2020

type: system collection: sok-references status: auto

Links: DOI · arXiv:1908.06972

Summary

Auto-extracted summary (unreviewed)

PrivFT performs private text classification with CKKS FHE in two modes: inference on encrypted user inputs with a plaintext fastText-style model (no accuracy loss), and training an encrypted model on a fully encrypted dataset. The authors implement an RNS variant of CKKS on GPUs (built on the A*HE library), obtaining 1-2 orders of magnitude speed-up over the CPU SEAL implementation, with under 0.66 s per encrypted inference on GPU, while encrypted training takes 5.04 days on GPU (11.1 days on CPU).

Abstract

The need for privacy-preserving analytics is higher than ever due to the severity of privacy risks and to comply with new privacy regulations leading to an amplified interest in privacy-preserving techniques that try to balance between privacy and utility. In this work, we present an efficient method for Text Classification while preserving the privacy of the content using Fully Homomorphic Encryption (FHE). Our system (named \textbf{Priv}ate \textbf{F}ast \textbf{T}ext (PrivFT)) performs two tasks: 1) making inference of encrypted user inputs using a plaintext model and 2) training an effective model using an encrypted dataset. For inference, we train a supervised model and outline a system for homomorphic inference on encrypted user inputs with zero loss to prediction accuracy. In the second part, we show how to train a model using fully encrypted data to generate an encrypted model. We provide a GPU implementation of the Cheon-Kim-Kim-Song (CKKS) FHE scheme and compare it with existing CPU implementations to achieve 1 to 2 orders of magnitude speedup at various parameter settings. We implement PrivFT in GPUs to achieve a run time per inference of less than 0.66 seconds. Training on a relatively large encrypted dataset is more computationally intensive requiring 5.04 days.
Source: arxiv

Extracted details

Field Value Source Confidence
contribution PrivFT system for homomorphic text-classification inference with zero accuracy loss llm 0.80
contribution homomorphic training of a text-classification model on fully encrypted data llm 0.80
contribution GPU (CUDA) implementation of RNS-CKKS with 1-2 orders of magnitude speed-up over CPU SEAL llm 0.80
contribution CPU (SEAL) and GPU implementations with parallelised vector-matrix operations llm 0.80
schemes CKKS llm 0.80
libraries SEAL llm 0.80
libraries A*HE llm 0.80
libraries CUDA llm 0.80
libraries fastText llm 0.80
models fastText-style text classifier (embedding + linear layer) llm 0.80
tasks text classification llm 0.80
tasks sentiment analysis llm 0.80
tasks spam detection llm 0.80
tasks topic classification llm 0.80
tasks document classification llm 0.80
hardware NVIDIA GPU llm 0.80
hardware CPU llm 0.80
threat model untrusted cloud evaluator, client-server (MLaaS) llm 0.80
interaction non-interactive FHE llm 0.80
limitations homomorphic training is computationally intensive (days of runtime) llm 0.80
limitations GPU memory is limited, requiring tiling of the training computation llm 0.80
llm relevance medium llm 0.80
key result inference latency: <0.66 s (per encrypted inference on GPU across datasets) llm 0.80
key result GPU CKKS speed-up: 1-2 orders of magnitude (GPU RNS-CKKS vs CPU SEAL at various parameter settings) llm 0.80
key result encrypted training time: 5.04 days (GPU), 11.1 days (CPU) (training on a relatively large encrypted dataset) llm 0.80
key result accuracy loss: zero (homomorphic inference vs plaintext model) llm 0.80

SoK reference [8]

A. Al Badawi, L. Hoang, C. F. Mun, K. Laine, and K. M. M. Aung, “Privft: Private and fast text classification with homomorphic encryption,” IEEE Access, vol. 8, pp. 226 544–226 556, 2020.

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: CKKS 46 [p.1], MPC 6 [p.3], BFV 3 [p.8], Leveled / somewhat HE 2 [p.2]
  • library: Microsoft SEAL 17 [p.2], HElib 1 [p.15], Concrete (Zama) 1 [p.9]
  • model: RNN / LSTM 5 [p.4], BERT 4 [p.4], Transformer 2 [p.4], Logistic regression 2 [p.3], CNN 1 [p.14]
  • dataset: IMDB 4 [p.10], Yelp 4 [p.10], AG News 3 [p.10], ImageNet 1 [p.13], UCI repository 1 [p.11]
  • hardware: GPU 57 [p.1], Multi-GPU 5 [p.2], CUDA 4 [p.8], NVIDIA V100 3 [p.2], Intel Xeon 1 [p.10]
  • technique: RNS 23 [p.2], Rotations 16 [p.6], Rescaling 11 [p.5], Softmax 8 [p.5], Encrypted training 6 [p.2], Bootstrapping 4 [p.4], Matrix multiplication 4 [p.5], SIMD packing 3 [p.6], Key switching 2 [p.6], Relinearization 2 [p.5], Polynomial approximation 2 [p.8], Semi-honest model 2 [p.3]
  • assumption: LWE 3 [p.11]
  • task: Medical / healthcare 5 [p.1], Sentiment analysis 3 [p.4], Genomics 2 [p.1]
  • topic: Homomorphic encryption 83 [p.1], Benchmark 4 [p.2]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/sok-references/008_privft-private-and-fast-text-classification-with-homomorphic.pdf sok-references 16 1.4 MB 29c86f08fd43

Database key: privft-private-and-fast-text-classification-with-homomorphic-encryption · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set privft-private-and-fast-text-classification-with-homomorphic-encryption COLUMN VALUE