UC Irvine

Various in-situ and mobile sensors are deployed in the smart environments for the purpose of providing personalized services to the end users. Large amount of data is being collected, stored and analyzed from a variety of sensors in a real-time manner. Such data introduces several challenges to the end users. These challenges include privacy and security of the users and system development challenges such as scalability, data management, energy efficiency, and data analytics.

The functionality of a smart environment depends heavily on the context. For example, the HVAC (Heating, Ventilation, and Air Conditioning) system of a building takes into account the occupancy of various regions of the building and determines the appropriate temperature of the regions. Such decision requires determination of accurate contexts such as occupancy, air pressure, outside air temperature, and water pressure. However, the state of a motion detector used to derive occupancy can reveal user-privacy, due to revealing the absence or presence of a user at a given space. Similarly, participatory thermal comfort control systems take users thermal comfort votes into account to control the temperature of a building. These systems require fairness in decision making. For example, if the majority of the votes for their comfort level is “good” even though one person is uncomfortable, the IoT-enabled building management systems will keep ignoring the minor’s opinion. When applying decision-making methods to aggregate user opinions that is collected through sensor data, the result can be unfair to certain groups of users. Likewise, context-aware messages can benefit from taking user and environmental context into account to deliver messages in real time/near real time manner. Such a messaging application requires scalable context collection and evaluation of context predicates. However, enrichment techniques which convert low-level sensor readings into meaningful context cannot be applied to the context data on the resource-constrained mobile devices due to the high resource requirements of these algorithms.

We identify aforementioned context reasoning challenges within two different IoT testbeds: first, TIPPERS (Testbed for IoT-based Privacy-preserving PERvasive Spaces) is an experimental six-story smart building testbed in UC Irvine designed to study the numerous privacy challenges due to fine-grained monitoring of building occupants and visitors using a diverse set of sensors. Second, Honeywell testbed consists of total 7687 Enlighted sensors which are capable of recording occupancy, light, power, and temperature.

This thesis is organized as follows. First, we study occupancy sensors and propose a privacy attack in which the adversary associates an individual with the occupancy sensor by combining the occupancy data with other public information that could easily be obtained online, and that breaches the user-privacy. Second, we present the first study, up to the authors' knowledge, on fairness in the aggregation of user thermal comfort in Participatory Thermal Comfort Control (PCC) systems. Third, we propose a context-aware messaging framework, SCARF that empowers senders and receivers to control message delivery through policies defined over context collected from a variety of mobile and in-situ sensors.