Data-driven modeling of phenomena in wireless sensor networks
Wireless sensor networks (WSNs) is an emerging field with applications that span scientific, engineering, medical and other disciplines. Significant human and capital infrastructure is needed for testing behavior of WSNs in real-world deployments. Sensor network simulations have the advantage of facilitating testing without going through the rigors of a deployment, but require good simulation models. However, high-quality communication and phenomena models are extremely hard to come by and suffer from need of large quantity of training data to capture the time-varying nature of the underlying phenomenon. In my dissertation, I address issues in the modeling of wireless links (communication-related) and occupancy monitoring (application phenomenon-related). For wireless link modeling, I advocate a novel machine learning-based, data-driven approach involving Hidden Markov Models (HMMs) and Mixtures of Multivariate Bernoullis (MMBs) for modeling the long and short time scale behavior of links. For occupancy modeling, I propose SCOPES, a wireless camera sensor network for gathering building occupancy traces for aiding model creation. However, both problems require large quantity of training data to model the time varying nature of the underlying phenomenon. For each instance, I solve the training data problem by using parameter-tying based model adaptation techniques that constrain the new model parameters through a nonlinear transformation of a pre-existing reference model. Using model adaptation, I showed that we can achieve significant reduction in training data requirement, thereby improving the simulation quality by enabling the construction of high-quality models.