Technical Report 030040, Computer Science Department, University of California, Los Angeles, 2003. Paper abstract bibtex
Information theory had a broad, deep, and profound impact both in science and engineering. It was based on a number of new fundamental concepts. Of course, the proposed misinformation theory does not have those types of new and revolutionary concepts. In a sense, it is unlikely that any future theory related to semantic processing may have the simplicity and elegance of the theory and techniques related to syntax-only processing. However, we believe that misinformation theory can provide sound foundations and concepts to address issues related to the integrity of obtained and employed information and that it will have a broad spectrum of important applications. The theory can be generalized to address under-constrained problems by requiring that after the misinformation, none of the initial constraints be violated while the solution space is skewed to the maximal extent in accordance with proper misinformation metrics. While the theory is generic and can be applied to many domains (e.g. information posted on the Internet and in social networks), we demonstrate it in the context of wireless embedded sensor networks. The typical question that we address is what value a node should report in such a way that will maximally skew the conclusion of the user while not being detected as an attacker and not crossing a threshold where the user starts to suspect the obtained results. The misinformation problem is formulated using the canonical form for stating optimization problems (consisting of an objective function and constraints) and solved under a variety of assumptions about how the attacking nodes are detected, when the result is accepted as correct, protocols for information exchange, and knowledge of attacking sources. We begin by analyzing the simple scenario of single value reading and then conduct extensive simulations for trilateration and show under which conditions which type of misinforming is most effective.