Dynamic Data Driven Applications Systems



Dynamic Data Driven Applications Systems ("DDDAS") is a paradigm whereby the computation and instrumentation aspects of an application system are dynamically integrated with a feedback control loop, in the sense that instrumentation data can be dynamically incorporated into the executing model of the application (in targeted parts of the phase-space of the problem to either replace parts of the computation to speed-up the modeling or to make the model more accurate for aspects of the system not well represented by the model; this can be considered as the model "learning" from such dynamic data inputs), and in reverse the executing model can control the system's instrumentation to cognizantly and adaptively acquire additional data (or search through archival data), which in-turn can improve or speedup the model (modeling process). DDDAS-based approaches have been shown that they can enable more accurate and faster modeling and analysis of the characteristics and behaviors of a system and can exploit data in intelligent ways to convert them to new capabilities, including decision support systems with the accuracy of full-scale modeling, executing model-driven adaptive management of complex instrumentation (including adaptive coordination across multitudes of heterogeneous sensors and controllers), as well as efficient data collection, management, and data mining.

The power of the DDDAS paradigm is that it involves a dynamically adapting and system-cognizant model (for example a model cognizant of the physics of the system, or other inherent characteristics and representations of the system), which "learns" and adapts through the "dynamic data" inputs at execution time, can discern false data and avoids the pitfalls of traditional Machine Learning approaches which can go rogue. Moreover, unlike ML methods, DDDAS enables more accurate and faster modeling and analysis, for "systems analytics" rather than simply "data analytics", and the DDDAS computational and instrumentation frameworks, include in addition to comprehensive system-characteristics cognizant representations and models, software and hardware (computational and instrumentation) platforms architectures and services, and can also include the human-in-the-loop, as complex systems typically involve.

DDDAS-based approaches have demonstrated new capabilities in systems modeling and instrumentation, as well as autonomic capabilities in many areas, ranging from fundamental studies in materials properties (e.g., nanomaterials), to structural and civil engineering (e.g., smart buildings) and aerospace, to manufacturing (process planning and control; additive manufacturing), transportation systems, energy systems (e.g., smart powergrids), environmental (e.g., wildfires), weather (atmospheric and space), medical diagnosis and treatment, cloud computing, IoT, and communications systems, cybersecurity, and more. The DDDAS site contains links on the extensive work and impact of the DDDAS paradigm.

History
The DDDAS concept - and the term - was proposed by Frederica Darema, starting in the early 1980s and through the 1990s,  who initiated the efforts within the National Science Foundation (NSF) and led the organization of a workshop in March 2000, where she designated as academic co-Chairs of the workshop, Profs Craig Douglas and Abhi Deshmukh. Around 2008, Darema introduced the term Infosymbiotics or Infosymbiotic Systems to denote DDDAS. Many researchers in academia, industry, and labs were influenced and adopted the DDDAS concept and the term, and conducted research under Dr. Darema's programs, starting from the mid-1990', at DARPA, NSF (including multi-agency programs), and AFOSR, and continued by Dr. Blasch after he took over as Program Manager at AFOSR, upon Dr. Darema becoming the Director of AFOSR in 2016. Thus, a community has formed and advanced systems capabilities and related new concepts under the rubric of DDDAS.

Starting in 2000 Dr. Darema led the community in organizing a number of DDDAS forums; these include: a series of DDDAS Workshops, Symposia, Panels, and other related activities, for example in conjunction with the International Conference in Computational Sciences (ICCS), the International Parallel and Distributed Computing Symposium (IPDPS), the Winter Simulation Conferences (WSC), the American Controls Conference (ACC), and more. Since 2016, Dr. Blasch also has organized numerous DDDAS and other related forums (e.g., Fusion2015 and follow-up Conference series). In 2014 an environmentally focused DDDAS-based conference, DyDESS 2014 was held at MIT, organized by Dr Ravela. The DDDAS2016 conference was sponsored by United Technologies in Hartford CT, and commenced an international conference series, with follow-up in 2017, 2020, 2022 hosted at MIT, and DDDAS2024 hosted at Rutgers University (with more planned in the future), with conference proceedings published by Springer. Other work is presented in the DDDAS Handbook series by Springer. A more complete list of DDDAS forums and other activities is provided in the DDDAS site.

Related concepts
DDDAS-based approaches advance the state of the art over a number of somewhat related but more limited concepts, which have been proposed over preceding years, and which are subsets of the more comprehensive and powerful DDDAS paradigm of feedback-control between an executing model of a system with its instrumentation. Below are some indicative examples.


 * Learning methods (starting in the 1980s and 1990s) and concepts like Reinforcement Learning (starting in the mid-1990s), which has the "flavor" of DDDAS, nevertheless, in these approaches there is no underlying system-cognizant, comprehensive model of the system(application) under study, nor there is control of the system instrumentation, like in DDDAS.
 * The data assimilation concept where observation data are used to correct and constrain uncertainty in computed data-points (computed vector of a physical parameter in the model) is more limited than the DDDAS concept where the dynamic data inputs can replace a patch of the mesh and for multiple parameters. Moreover, the reverse aspect in the DDDAS feedback control loop – that is the model adaptively controlling the instrumentation, this was also adopted later by the Data Assimilation concept, for example, the Data Assimilation and Adaptive Observation, MIT Thesis in 1999, which discusses advances to traditional data assimilation, the ideas presented there on integrated supplementary targeted observations into the assimilation system using forecasts and their uncertainties - the approaches discussed there-in are static, akin to human-in-the loop computational steering, applied to the observational aspect.
 * DDDAS approaches in controlling instrumentation are beyond the traditional ideas presented in the late 1950s to 1970s, such as by Chertoff on Sequential Design of Experiments, and by Fedorov on Design of Experiments (1970s), where the experiments determine the parameters and the parameters guide the experiments, including for data and model selection - these approaches are the classical serialized approaches of the human in the loop determining statically what models to use and what experiments to conduct and repeat the process. MacKay's Information-based Active Data Selection (1991) employs Bayesian methods to determine expected informativeness of candidate measurements is used to select salient ones for learning, improving the expected informativeness. And, Information Retrieval (in the 90s), where queries generate searches, and the results refine the queries with relevance feedback.