Inductive modeling of discrete event systems: A TMS-based non-monotonic reasoning approach.
Committee ChairCellier, Francois
MetadataShow full item record
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
AbstractIn this presentation, we propose a framework for inductive modeling of discrete-event systems called Discrete-event Inductive Reasoner, DIR. It is based on systems theory and non-monotonic logic. We present a new representation for a finite set of discrete-event observed input/output time segments called Iterative IOFO (Input/Output Function Observation) Specification. We also introduce a novel use of non-monotonic logic allowing DIR to make tentative decisions. With the inclusion of additional data, non-monotonic logic ensures that any of its prior decisions that becomes violated is retracted properly. Due to the underlying features of non-monotonic reasoning, DIR supports incremental refinement/extension of the model iterative IOFO specification. To implement the DIR, we map the iterative IOFO into a logic-based representation suitable for Logic-based Truth Maintenance System, a form of non-monotonic reasoning mechanism. Abstraction mechanisms are defined that are capable of predicting unobserved input/output time segments, given some existing IO segments and some assumptions. The systems theory framework enables us to develop the means to ensure the appropriate use of abstractions. In this way, the model is incrementally extended by predicting and retaining unobserved IO segments in a well-defined fashion. Also, we discuss an implemented prototype of DIR called Logic-based Discrete-event Inductive Reasoner, LDIR. Two examples are used to discuss LDIR's features. We give some heuristic metrics for quantitative evaluation of LDIR's predictions. We provide general guidelines for the evaluation of DIR and place the methodology within existing inductive modeling approaches. We conclude with some shortcomings of our approach and speculate on future research directions.
Degree ProgramElectrical and Computer Engineering