Sense and respond systems are employed in diverse applications. The Infospheres group is doing research on a general theory of S&R systems. Whether the application is interdicting radiological weapons, managing a stock portfolio, or helping disabled people live independently in their own homes, the overall system architecture and design considerations have many points in common. Our research is a contribution towards the development of a discipline of S&R systems. This research includes a study of prior and ongoing work in several disciplines including concurrent computing, control theory, signal processing, statistics, optimization and game theory.
Metrics for Evaluating S&R Systems
Theories require measures for comparing alternatives. Metrics, such as computational complexity are inadequate for comparing S&R systems. This research proposes quantitative metrics for S&R systems and develops algorithms for evaluating them. The design of S&R systems is framed as an optimization problem where the objective function is defined in terms of these metrics. This research helps software architects analyze the costs and benefits of S&R applications in a systematic way.
Design Patterns for S&R Systems
S&R systems consist of components: sensors, communication links, processors and responders. Humans may play many of these roles. Many varieties of components with different costs and capabilities are available. Given budgets, overall goals (objective functions) and constraints such as maximum tolerable error rates, this research investigates optimum designs and optimum operations. Optimal designs consist of creating the best decisions during design time including selections of components, determining their locations and connectivity, and developing algorithms for optimally sensing and responding to the environment.
Event Detection and Prediction
An event is a significant change in the state of an organization or its environment. The change may occur rapidly or it may be a consequence of many incremental changes over a period of time. The detection of events with low rates of false positives (detection of nonevents) and false negatives (not detecting genuine events) is one of the challenges of designing S&R systems. Predicting events so that responses can be initiated early is particularly important in applications where rapid response is required. This research deals with statistics, signal processing, sensor networks and optimization.
Systems for Specifying How to Sense and Respond
Higher biological organisms, including humans, learn what to sense and how to respond. This research investigates S&R platforms – i.e., general purpose S&R engines – that can be tailored by end-users and systems administrators to serve the specific needs of each organization. Tailoring a general-purpose S&R platform for a specific application requires mechanisms for specifying what is desired: what is to be sensed, and what responses are appropriate for what situations? Our research in this area investigates notations and explores machine-learning algorithms for specifying events and responses.
Formal Methods, Verification and Validation
Reasoning about the behavior of S&R systems falls into two categories. The first category deals with the questions: given a system and given enough time, will the system eventually converge to the desired solution, and is the desired solution stable? The second category deals with timeliness in response: how quickly will it get to the desired solution? The answer to the first type of question is either yes or no, whereas the answer to the second type is a continuous measure with shorter delays being better than longer ones. Work on program correctness answered the first type of question by proving correct program termination – the program will get to a desired state in finite time – whereas S&R deals with eventual convergence – prove that the system will get arbitrarily close to a desired state. This research integrates logics used in computer science to reason about program correctness with mathematics used in control and dynamical systems dealing with convergence and stability.