GRAPH-BASED LEARNING AND DECISION MAKING IN INFORMATION NETWORKS

Abstract

In this thesis, we address challenges regarding both information collection and information processing in Information Networks. The first half of the thesis discusses strategic information collection policies, from a crowdsourcing perspective, with an objective to get the most information with the lowest cost. Efficient information collection policies are proposed to achieve optimal system-wide utility while offering incentives to participants in stochastic crowdsourcing systems. The second half of the thesis discusses how to allocate learning resources in Information Networks to achieve efficient decision making. Using graphical models as data representations, we propose an algorithm to optimize the graphical model structure of sensor networks to obtain the best trade-off between learning performance and energy consumption. Furthermore, an active learning algorithm is proposed for graphical models to learn a good model with a smaller sample size.