Privacy preserving deep learning framework in fog computing
Document typeConference report
Rights accessOpen Access
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder
Nowadays, the widespread use of mobile devices has raised serious cybersecurity challenges. Mobile services and applications use deep learning (DL) models for the modelling, classification and recognition of complex data, such as images, audio, video or text. Users benefit from the wide range of ser-vices and applications offered by these devices but pay an enormous price, the privacy of their personal data. Mobile services collect all different types of us-ers’ data, including sensitive personal data, photos, videos, clinical data, bank-ing data, etc. All this data is pooled to the Cloud to train global DL models, and big companies benefit from all the collected users’ data, posing obvious serious privacy issues. This paper proposes a privacy preserving framework for Fog computing envi-ronments, which adopts a distributed deep learning approach. Internet of Things (IoT) end nodes never reveals their sensitivity to the Cloud server; instead, they share a fraction of the users’ data, blurred with Gaussian noise, with a nearby Fog node. The DL methods considered in this work are Multilayer Perceptron (MLP) and Convolutional Neural Networks (CNN), and for both cases the ac-curacy is similar to the centralized and privacy violating approach, obtaining the best results for the CNN model.
CitationGutiérrez, N. [et al.]. Privacy preserving deep learning framework in fog computing. A: International Conference on Machine Learning, Optimization, and Data Science. "Machine Learning, Optimization, and Data Science, 6th International Conference, LOD 2020: Siena, Italy, July 19-23, 2020: revised selected papers, part I". Berlín: Springer, 2020, p. 504-515. ISBN 978-3-030-64583-0. DOI 10.1007/978-3-030-64583-0_45.