A Context-aware Learning, Prediction And Mediation Framework For Resource Management In Smart Pervasive Environments
MetadataShow full item record
Advances in smart devices, mobile wireless communications, sensor networks, pervasive computing, machine learning, middleware and agent technologies, and human computer interfaces have made the dream of smart environments a reality. An important characteristic of such an intelligent, ubiquitous computing and communication paradigm lies in the autonomous and pro-active interaction of smart devices used for determining inhabitants' important contexts such as current and near-future locations, activities or vital signs. `Context Awareness' is perhaps the most salient feature of such an intelligent computing environment. An inhabitant's mobility and activities play a significant role in defining his contexts in and around the home. Although there exists optimal algorithm for location and activity tracking of a single inhabitant, the correlation and dependence between multiple inhabitants' contexts within the same environment make the location and activity tracking more challenging. In this thesis, first we propose a cooperative reinforcement learning policy for location-aware resource management in multi-inhabitant smart homes. This approach adapts to the uncertainty of multiple inhabitants' locations and most likely routes, by varying the learning rate parameters. Using the proposed cooperative game-theory based framework, all the inhabitants currently present in the house attempt to minimize this overall uncertainty in the form of utility functions associated with them. Joint optimization of the utility function corresponds to the convergence to Nash equilibrium and helps in accurate prediction of inhabitants' future locations and activities. Hypothesizing that every inhabitant wants to satisfy his own preferences about activities, next we look into the problem from the perspective of non-cooperative game theory where the inhabitants are the players and their activities are the strategies of the game. We prove that the optimal location prediction across multiple inhabitants in smart homes is an NP-hard problem and to capture the correlation and interactions between different inhabitants' movements (and hence activities), we develop a novel framework based on a non-cooperative game theoretic, Nash H-learning approach that attempts to minimize the joint location uncertainty of inhabitants. Our framework achieves a Nash equilibrium such that no inhabitant is given preference over others. This results in more accurate prediction of contexts and more adaptive control of automated devices, thus leading to a mobility-aware resource (say, energy) management scheme in multi-inhabitant smart homes. Experimental results demonstrate that the proposed framework is capable of adaptively controlling a smart environment, significantly reduces energy consumption and enhances the comfort of the inhabitants. To promote independent living and wellness management services in this smart home environment we envision sensor rich computing and networking environments that can capture various types of contexts of patients (or inhabitants of the environment), such as their location, activities and vital signs. However, in reality, both sensed and interpreted contexts may often be ambiguous, leading to fatal decisions if not properly handled. Thus, a significant challenge facing the development of realistic and deployable context-aware services for healthcare applications is the ability to deal with ambiguous contexts to prevent hazardous situations. In this thesis, we propose a quality assured context mediation framework, based on efficient context-aware data fusion and information theoretic system parameter selection for optimal state estimation in resource constrained sensor network. The proposed framework provides a systematic approach based on dynamic Bayesian network to derive context fragments and deal with context ambiguity or error in a probabilistic manner. It has the ability to incorporate context representation according to the applications' quality requirement. Experimental results demonstrate that the proposed framework is capable of choosing a set of sensors corresponding to the most economically efficient disambiguation action and successfully sensing, mediating and predicting the patients' context state and situation. Energy-efficient determination of an individual's context (both physiological and activity) is an important technical challenge for this assisted living environments. Given the expected availability of multiple sensors, context determination is viewed as an estimation problem over multiple sensor data streams. We develop a formal, and practically applicable, model to capture the tradeoff between the accuracy of context estimation and the communication overheads of sensing. In particular, we propose the use of tolerance ranges to reduce an individual sensor's reporting frequency, while ensuring acceptable accuracy of the derived context. We introduce an optimization technique allowing the context service to compute both the best set of sensors, and their associated tolerance values, that satisfy the QoINF target at minimum communication cost. Experimental results with SunSPOT sensors are presented to attest to the promise of this approach.