상세 보기
Privacy-Preserving Traffic Flow Prediction: A Split Learning Approach
- Tran, N.-P.;
- Dao, N.-N.;
- Do, Q.-T.;
- Nguyen, T.-V.;
- Cho, Sungrae
WEB OF SCIENCE
5SCOPUS
8초록
Nowadays, a crucial component of Intelligent Transportation Systems is traffic flow prediction, contributing to reducing traffic congestion or traffic accidents. Many techniques are designed to exploit the advantages of Artificial Intelligence (AI) in traffic flow prediction. Machine Learning enjoys the benefits of diverse transportation data, which is collected by organizations, private companies, and governments. On the other hand, these activities raise considerable concerns about privacy because these data might contain confidential information. To deal with this challenge, we provide a unique machine-learning method for predicting traffic flow, that jointly considers data privacy and communication costs. In this work, a Gated Recurrent Units (GRUs) neural network is designed in the Split Learning (SL) setting for the traffic flow prediction problem. The clients in the centralized learning approach need to exchange their data with a central server. In contrast, SL respects personal privacy, it allows clients and servers collaboratively train deep neural networks without passing raw data between clients and the main server. They only exchange activation and gradients in the training phase, so data privacy is guaranteed. Furthermore, we equip Autoencoder as a compression tool to reduce the size of the client's activation before sending it to the central server, it reduces communication overhead in training time. © 2023 IEEE.
키워드
- 제목
- Privacy-Preserving Traffic Flow Prediction: A Split Learning Approach
- 저자
- Tran, N.-P.; Dao, N.-N.; Do, Q.-T.; Nguyen, T.-V.; Cho, Sungrae
- 발행일
- 2023-01
- 유형
- Proceedings Paper
- 저널명
- International Conference on Information Networking
- 권
- 2023-January
- 페이지
- 248 ~ 250
- 언어
- ENG
- 출판사
- IEEE Computer Society
- 분량
- 3 페이지
- ISSN
- P 1976-7684