مشخصات مقاله:
عنوان فارسی مقاله:
یادگیری عمیق چند جانبه با افزایش حریم خصوصی
عنوان انگلیسی مقاله:
Privacy-enhanced multi-party deep learning
کلمات کلیدی مقاله:
حریم خصوصی، یادگیری عمیق چند جانبه، حریم خصوصی دیفرانسیلف، رمزگذاری Homomorphic، بودجه حفظ حریم خصوصی
کلمات کلیدی انگلیسی:
Privacy – Multi-party deep learning – Differential privacy – Homomorphic encryption – Privacy budget
مناسب برای رشته های دانشگاهی زیر:
مهندسی کامپیوتر
مناسب برای گرایش های دانشگاهی زیر:
مهندسی الگوریتم و محاسبات،هوش مصنوعی، امنیت اطلاعات
وضعیت مقاله انگلیسی و ترجمه:
مقاله انگلیسی را میتوانید به صورت رایگان با فرمت PDF با کلیک بر روی دکمه آبی، دانلود نمایید. برای ثبت سفارش ترجمه نیز روی دکلمه قرمز رنگ کلیک نمایید. سفارش ترجمه نیازمند زمان بوده و ترجمه این مقاله آماده نمیباشد و پس از اتمام ترجمه، فایل ورد تایپ شده قابل دانلود خواهد بود.
فهرست مطالب:
Abstract
Keywords
1. Introduction
2. Preliminaries
2.1. Deep learning
2.1.1. Distributed stochastic gradient descent
2.2. Attack model
2.3. Differential privacy
2.4. Homomorphic encryption
3. Our framework
3.1. Overview
3.2. Participants
3.2.1. Homomorphic encryption
3.2.2. Strategies for allocating privacy budget
3.3. Central server
3.4. Analysis
3.4.1. Effectiveness
3.4.2. Efficiency
3.4.3. Privacy
4. Experimental evaluation
4.1. Datasets
4.2. Experiments setup
4.3. Experiments results
4.3.1. Results for low noise training and high noise training
4.3.2. Results for strategies for allocating privacy budget
4.3.3. Results for homomorphic encryption
4.3.4. Parameter analysis
5. Conclusion and future work
Acknowledgments
References
قسمتی از مقاله انگلیسی:
1. Introduction
In recent years, deep learning has achieved impressive success in many fields (e.g., computer vision Camuñas-Mesa, SerranoGotarredona, Ieng, Benosman, & Linares-Barranco, 2018, natural language processing Devlin, Chang, Lee, & Toutanova, 2018, speech processing Zhang, Li, Jin, & Choe, 2015 and human -machine games Silver et al., 2016, etc.), and its performance has approached or even exceeded human in a wide range of applications, such as in face recognition (Cao, Wang, Gao, & Li, 2018; Huang, Dai, Ren, & Lai, 2017), handwritten digit recognition (Liang & Hu, 2015; Wang, Xu, Yang, & Zurada, 2018) and playing Go (Silver et al., 2017). A prerequisite for these successes is the massive data available for model training. However, massive data collection from multiple sources may present privacy issues. On the one hand, data may contain both identity information and private information. For example, a photo contains both faces and elements which may suggest religious beliefs. On the other hand, it is illegal to share sensitive data, like medical data stored by medical institutions. Once these kinds of data are collected centrally, the data may be stored permanently, and used for the purpose which the original data owner is unaware of. Generally, these private or sensitive data are scattered across multiple research institutions or companies, and they are unwilling or unable to share data with each other. If researchers want to train a deep learning model with high accuracy, they can only perform on their own data. Nevertheless, data owned by an institution may be very limited (e.g., a school clinic), which will result in an over-fitting deep learning model.