Private and Scalable Personal Data Analytics Using Hybrid Edge-to-Cloud Deep Learning

Abstract

We are observing an increasing presence of cyber-physical systems and their associated data around us. While the ability to collect, collate, and analyze the vast amount of rich information from smartphones, IoT devices, and urban sensors can be beneficial to the users and the industry, this process has led to a number of challenges ranging from performing efficient and meaningful analytics on the generated big data, to privacy challenges associated with the inferences made from these data due to ubiquitous nature of connected devices. In this paper, we discuss novel edge-computing methods to improve the scalability and privacy of user-centered analytics. We present a hybrid framework where edge devices and resources centered around the user can complement the cloud for providing privacy-aware, yet accurate and efficient analytics. We present early evaluations of the proposed framework on a number of exemplar applications, and discuss the broader implications of such approaches.

A review of privacy-preserving techniques for deep learning

“…As for efficiency, the obfuscation introduces a time overhead proportional to the input size, considered by the authors as acceptable for real-time processing. Osia et al [39,67,68] described a hybrid architecture for inference based on model splitting concept. The layers are split between local device and the cloud, and the process into feature extraction, performed by the local primary layers, and analysis, performed by secondary layers on the cloud.…”

Toward Intelligent Industrial Informatics: A Review of Current Developments and Future Directions of Artificial Intelligence in Industrial Applications

“…In [51], an overview of algorithmic and processor techniques for transitioning deep learning to IoT and CPS is provided. A hybrid framework for privacy-preserving, accurate, and efficient analytics of IoT and CPS data based on user-centered edge devices and cloud computing is proposed in [52]. In [20], authors attempt to use industrial IoT to detect anomalies in power consumption based on a multi-view stacking intelligent ensemble.…”

Deep Learning for Privacy in Multimedia

“…In particular, we show how to generate natural-looking adversarial images either by selectively modifying colors within chosen ranges that we perceive as natural or by enhancing details in the image. The references that will be covered in the tutorial are [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18].…”