Privacy-Preserving Transformers: SwiftKey's Differential Privacy Implementation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Abouelenin, Abdelrahman, Abdelrehim, Mohamed, Fahim, Raffy, Hendy, Amr, Afify, Mohamed
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916728375607296
author Abouelenin, Abdelrahman
Abdelrehim, Mohamed
Fahim, Raffy
Hendy, Amr
Afify, Mohamed
author_facet Abouelenin, Abdelrahman
Abdelrehim, Mohamed
Fahim, Raffy
Hendy, Amr
Afify, Mohamed
contents In this paper we train a transformer using differential privacy (DP) for language modeling in SwiftKey. We run multiple experiments to balance the trade-off between the model size, run-time speed and accuracy. We show that we get small and consistent gains in the next-word-prediction and accuracy with graceful increase in memory and speed compared to the production GRU. This is obtained by scaling down a GPT2 architecture to fit the required size and a two stage training process that builds a seed model on general data and DP finetunes it on typing data. The transformer is integrated using ONNX offering both flexibility and efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2505_05648
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Privacy-Preserving Transformers: SwiftKey's Differential Privacy Implementation
Abouelenin, Abdelrahman
Abdelrehim, Mohamed
Fahim, Raffy
Hendy, Amr
Afify, Mohamed
Computation and Language
Cryptography and Security
Machine Learning
In this paper we train a transformer using differential privacy (DP) for language modeling in SwiftKey. We run multiple experiments to balance the trade-off between the model size, run-time speed and accuracy. We show that we get small and consistent gains in the next-word-prediction and accuracy with graceful increase in memory and speed compared to the production GRU. This is obtained by scaling down a GPT2 architecture to fit the required size and a two stage training process that builds a seed model on general data and DP finetunes it on typing data. The transformer is integrated using ONNX offering both flexibility and efficiency.
title Privacy-Preserving Transformers: SwiftKey's Differential Privacy Implementation
topic Computation and Language
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2505.05648