"We address one of the fundamental challenges of our time: Acting effectively while facing a deluge of data. Massive volumes of data are generated from corporate and public sources every second, in social, scientific and commercial applications. In addition, more and more low level sensor devices are becoming available and accessible, potentially to the benefit of myriads of applications. However, access to the data is limited, due to computational, bandwidth, power and other limitations. Crucially, simply gathering data is not enough: we need to make decisions based on the information we obtain. Thus, one of the key problems is: How can we obtain most decision-relevant information at minimum cost?Most existing techniques are either heuristics with no guarantees, or do not scale to large problems. We recently showed that many information gathering problems satisfy submodularity, an intuitive diminishing returns condition. Its exploitation allowed us to develop algorithms with strong guarantees and empirical performance. However, existing algorithms are limited: they cannot cope with dynamic phenomena that change over time, are inherently centralized and thus do not scale with modern, distributed computing paradigms. Perhaps most crucially, they have been designed with the focus of gathering data, but not for making decisions based on this data.We seek to substantially advance large-scale adaptive decision making under partial observability, by grounding it in the novel computational framework of adaptive submodular optimization. We will develop fundamentally new scalable techniques bridging statistical learning, combinatorial optimization, probabilistic inference and decision theory to overcome the limitations of existing methods. In addition to developing novel theory and algorithms, we will demonstrate the performance of our methods on challenging real world interdisciplinary problems in community sensing, information retrieval and computational sustainability."