Machine Learning Systems in the IoT: Trustworthiness Trade-offs for Edge Intelligence

Machine learning systems (MLSys) are emerging in the Internet of Things (IoT) to provision edge intelligence, which is paving our way towards the vision of ubiquitous intelligence. However, despite the maturity of machine learning systems and the IoT, we are facing severe challenges when integrating MLSys and IoT in practical context. For instance, many machine learning systems have been developed for large-scale production (e.g., cloud environments), but IoT introduces additional demands due to heterogeneous and resource-constrained devices and decentralized operation environment. To shed light on this convergence of MLSys and IoT, this paper analyzes the tradeoffs by covering the latest developments (up to 2020) on scaling and distributing ML across cloud, edge, and IoT devices. We position machine learning systems as a component of the IoT, and edge intelligence as a socio-technical system. On the challenges of designing trustworthy edge intelligence, we advocate a holistic design approach that takes multi-stakeholder concerns, design requirements and trade-offs into consideration, and highlight the future research opportunities in edge intelligence.

I. Introduction

Machine learning systems are omnipresent and tireless silent helpers that bring order to our busy modern life: they guide us through traffic, classify and predict diseases in humans and plants, and are our eyes and ears in situations where we cannot see and hear. Their underlying machinery are machine learning algorithms that fit complex functions over data to discover patterns and correlations which can be exploited to discover trends and relationships, and for making predictions [1]. Many machine learning algorithms can be scaled to very large datasets and improve with more data. This has made them extremely successful in analysing the large volumes of data produced by digital, online services and applications. Deep neural networks in particular have produced state of the art results for many perception based tasks and are now widely used to process image, video, speech, audio and sequential data [2]. Machine learning is a promising technique when a system or process is not well understood, or too complex and difficult to model explicitly, but data that can surface insights about it has been collected [3]. Equally, if applications are dynamic and evolve over time, machine learning systems can use new data to discover patterns and update their predictions, thus adapting with the application.