Intelligent IoT for the Digital World

5Lightweight Privacy‐Preserving Learning Schemes*

5.1 Introduction

With the advances of sensing and communication technologies, the internet of things (IoT) will become a main data generation infrastructure in the future. The drastically increasing amount of data generated by IoT will create unprecedented opportunities for various novel applications powered by machine learning (ML). However, various system challenges need to be addressed to implement the envisaged intelligent IoT.

IoT in nature is a distributed system consisting of heterogeneous nodes with distinct sensing, computing, and communication capabilities. Specifically, it consists of massive mote‐class sensors deeply embedded in the physical world, personal devices that move with people, widespread network fog nodes such as wireless access points, as well as the cloud backend. Implementing the fabric of IoT and ML faces the following two key challenges:

• Separation of data sources and ML computing power. Most IoT data will be generated by the end devices that often have limited computation resources, while the computing power needed by ML model training and execution will be located at the fog nodes and in the cloud. In addition, the communication channels between the end devices and the edge/cloud are often constrained, in that they are limited in bandwidth, intermittency, and have long delays.
• Privacy preservation. As the end devices can be deeply embedded in people's private space and time, the data generated ...