Modeling Dynamical Phenomena in the Era of Big Data

As the world around us gets equipped with widespread sensing, computing, communication, and actuation capabilities, opportunities to improve the quality of life arise. Smart infrastructures promise to dramatically increase safety and efficiency. While data abounds, the modeling and understanding of large-scale complex systems, such as energy distribution, transportation, or communication networks, water management systems, and buildings, presents several challenges. Deriving models from first principles via white or gray box modeling is infeasible. Classical black-box modeling is also not practical as model selection is hard, interactions change over time, and evolution can be observed passively without the chance to conduct experiments through data injection or manipulation of the system. Moreover, the causality structure of such systems is largely unknown. We contend that determining data-driven, minimalistic models, capable of explaining dynamical phenomena and tracking their validity over time, is an essential step toward building dependable systems. In this work we will outline challenges, review existing work, and propose future research directions.