The objective of this grant is to conduct a workshop that brings together leading researchers from around the world, specializing in mathematics, operations research, economics, computer science, engineering, cognitive sciences, psychology, and other fields, to discuss fundamental scientific issues in creating Self-Optimizing Systems (SOS). The goal of this workshop is to formulate a cross-disciplinary research agenda to develop fundamental understanding and create a knowledge base for the design and control of SOS that are robust to unexpected failures or unplanned events. This workshop is expected enhance the fundamental knowledge base for SOS by promoting key advances in theoretical foundations and self-optimization approaches.
If successful, this workshop will lead to the creation of a coherent scientific basis for research in SOS. This workshop will result in the development of a common set of theoretical frameworks, tool and technologies that would be available to a broad range of users to design next generation of engineered systems that can self-organize and perform efficiently in uncertain environments. These common principles and tools would allow coordinated self-organization at different levels of a system. This workshop will enhance our ability to address societal issues, for example managing large-scale infrastructure systems such as energy grid and water, and predicting natural-social phenomenon such as climate change, that result from complex interactions among large number of coupled natural and human sub-systems.
Today’s modern society is built on engineered systems, ranging from large-scale infrastructures, such as the power grid and transportation networks, to small-scale devices, such as MEMS motors for unclogging blood vessels and 3-D nano-ensembles for data storage, that need to function in an ever changing environment and under extreme uncertainty. The ability to design predictable, efficient and stable engineered systems is critical to ensuring safety, well-being, and competitiveness of the society. In recent years, Self-Optimizing Systems (SOS), defined as those systems which observe the environment, learn from the past actions/responses and adapt to the current situation in real time, have been developed to handle uncertainty in a variety of application domains. SOS reacts autonomously to changing environmental conditions. They can use their past experience to select the most appropriate action, or use a predictive model to simulate the future, or use a combination of both methods. They have the capability to learn and optimize their behavior during operation. Moreover, SOS are also able to reconfigure structurally, as well as parametrically based on the environment and their observations. The goal of the workshop funded under this grant was to formulate a cross-disciplinary research agenda to develop fundamental understanding and create a knowledge base for the design and control of SOS that are robust to unexpected failures or unplanned events. Key advances in theoretical foundations, such as mathematical formalisms for self-optimizing systems; and self-optimization approaches, such as optimal degree of adaptation, need to be made in order to develop a fundamental knowledge based to build SOS. The workshop was held from June 13-15, 2010 in Santa Fe, NM. The workshop had 22 attendees representing multiple disciplines, academic institutions and funding agencies. The workshop agenda consisted of plenary presentations and breakout sessions to discuss theoretical foundations, methods and approaches, and research challenges that need to be overcome to realize functional SOS. Some of the key findings of this workshop are listed below: • Although complexity and multi-agent structure may obfuscate the definition of SOS, the most general defining statements proposed were: (1) A system that self-organizes for the purpose of self-optimization (implicitly or explicitly); (2) Systems that change (potentially optimally) in response to environments. • Metrics for assessing SOS include the quality or normal performance of the system, how well the system can re-optimize when disturbed/perturbed, as well as the rate of convergence to equilibrium, and the robustness of the self-organized system. • Common training and techniques for studying SOS include optimization, probability theory, stochastic processes, game theory, control theory, dynamical systems, evolutionary theory, and empirical methods. • Methods and approaches identified for SOS: model checking methods for formalizing abstraction, design of hierarchy in architectures, adaptation of what agents control what variables, add/remove agents, optimization algorithms that learn by searching space changes over time in response to regularity, robust, dynamic optimization with time-dependent objectives, modeling human rationality via utility elicitation, explicit vs. intrinsic rewards, evolution of preferences, and learning how to play games. By bringing together experts from various disciplines, this workshop provided a setting with which we could start to formalize a knowledge base and develop a research agenda. The identified state of the art in SOS brings together techniques and models from academic disciplines of computer science, operations research, statistics, economics, sensors and control theory. Included in those techniques and models are learning algorithms, optimization, simulation, stochastic processes, approximation methods, equilibria concepts, and hierarchical modeling. Open research questions include the need to manage the tradeoffs between exploration and exploitation for multiple agents in an uncertain environment, and the fidelity and scalability of approximation methods when systems are too complex to be explicitly modeled. The workshop participants emphasized the range of work being done in the SOS area. There was a broad consensus that SOS can be a unifying paradigm for researchers and techniques at work in this area. The participants concurred that collaboration and networking opportunities established as a result of this workshop would prove valuable in advancing the science of SOS.
