UC Berkeley

The Internet of Things (IoT) is a term that represents a huge technological trend that is taking place: almost every device is being imbued with the intelligence of a microprocessor and an Internet connection. We view IoT as a phenomena in which new service models will emerge. Central to these service models will be the provided data and the conversations surrounding it. In this document, we outline our research in the formulation and analysis of these new service models. This work is focused on the role of data and the value of information in IoT.

First, we present our work on disaggregation algorithms, which take aggregate measurements at a higher level of abstraction to infer component measurements at a lower level of abstraction. This is inspired by many IoT settings where aggregation frequently happens along the data pipeline due to energy and bandwidth constraints, as well as limitations on sensor placement. Additionally, we present our work on blind system identification, which provide a method to identify the dynamics of observed systems when both the internal states and the inputs are not observed, as is typical in many IoT settings. For example, smart meters observe the aggregate energy consumption of a building, but do not directly observe the individual device's energy consumptions, the transient energy consumption dynamics of devices, or how devices are being utilized inside a building. Disaggregation and blind system identification allow IoT system managers to infer models of these components of IoT systems even when sensor measurements do not provide them.

Second, we present our work in quantifying, analyzing, and incentivizing privacy in IoT systems. Motivated in part by the efficacy of disaggregation algorithms, we consider the privacy facet of IoT technologies. We discuss the literature on quantifying privacy, and also discuss a new metric inspired by classical information theoretic and statistical frameworks, which we call {inferential privacy}. We translate some of the existing information theoretic and statistical literature into privacy guarantees in this new framework. Then, we discuss different design paradigms for privacy, which range from passive privacy analysis to optimal privacy-by-design. These taxonomies are complemented by detailed examples in transportation networks, smart grid control, and air quality regulation.

Third, we discuss the value of information, data markets, and new service models in IoT. We consider a model of a data buyer who deal with strategic data sources: the data buyer must balance its objective of having a low-error statistical estimator with the cost of issuing incentives to effort-averse data sources. We extend these models to competitive settings, where multiple data buyers are incentivizing the same data sources, and analyze the existence of equilibria in such settings, as well as properties of such equilibria. We also consider how IoT systems allow for a new mode of actuation: causal imputation. In many smart infrastructure applications, we no longer have control commands that directly affect the dynamics. For example, the turn signal on a car does not have any direct effect on the dynamics of the car. As another example, many demand response programs are moving towards preferential pricing, rebates, and other incentive schemes to curtail peak energy consumption, rather than a direct load control. Both of these situations can be modeled with our optimal causal imputation framework. These control actions are structurally different from previously studied modes of actuation and we formulate the problem of optimal causal imputation, and provide algorithms for calculating these solutions under certain assumptions.

In this document, we provide a technical analysis of the IoT systems and the statistical properties of their data, a behavioral analysis of the human actors who respond to IoT systems and participate by the revelation of their data (or lack thereof), and a game theoretic analysis of the data analytics companies who drive competitive data markets with market power. This work is an effort to build a larger picture of the IoT as an emerging data market, and motivates much of the theoretical frameworks we have developed and plan to develop in future work.