A Framework for Evidential-Reasoning Systems

Abstract

Evidential reasoning is a body of techniques that supports automated reasoning from evidence. It is based upon the Dempster-Shafer theory of belief functions. Both the formal basis and a framework for the implementation of automated reasoning systems based upon these techniques are presented. The formal and practical approaches are divided into four parts (1) specifying a set of distinct propositional spaces, each of which delimits a set of possible world situations (2) specifying the interrelationships among these propositional spaces (3) representing bodies of evidence as belief distributions over these propositional spaces and (4) establishing paths for the bodies of evidence to move through these propositional spaces by means of evidential operations, eventually converging on spaces where the target questions can be answered. I IntroductionFor the past several years, we have been addressing perceptual problems that bridge the gap between low-level sensing and high-level reasoning [Low82,GLF81,LG83b,LG83a,LSG86,Wes86].Problems that fall into this gap are often characterized by multiple evidential sources of real-time data, which must be propframework for reasoning with perceptual data that forms the basis for evidential-reasoning1 systems.The information required to understand the current state of the world comes from multiple sources: real-time sensor data, previously stored general knowledge, and current contextual information. Sensors typically provide evidence in support of certain conclusions.Evidence is characteristically uncertain:it allows for multiple possible explanations; it is incomplete:the source rarely has a full view of the situation ; and it may be completely or partially incorrect. The quality and the ease with which situational information may be extracted from a synthesis of current sensor data and prestored knowledge is a function both of how strongly the characteristics of the sensed data focus on appropriate intermediate conclusions and on the strength and effectiveness of the relations between those conclusions and situation events. Given its characteristics, evidence is not readily represented either by logical formalisms or by classical probabilistic estimates. Because of this, developers of automated systems that must reason from evidence have frequently turned to informal, heuristic methods for handling uncertain information.The "probabilities" produced by these informal approaches often cause difficulties in interpretation. The lack of a formally consistent method can cause problems in extending the capabilities of such systems effectively. Our work in evidential reasoning was motivated by these shortcomings.Our theory is based on the Shafer-Dempster theory of evidence [Dem68,Sha76,Sha86] and aims to overcome some of the difficulties in reasoning from evidence by providing a natural representation for evidential information, a formal basis for drawing conclusions from evidence, and a representation for belief.In evidential reasoning, a knowledge source (KS) is allowed to e...

An Extended Event Reasoning Framework for Decision Support under Uncertainty

“…In order to reflect the reliability of evidence, a Discount rate was introduced by which a mass function can be discounted [8]:…”

On the Orthogonal Projection of a Belief Function

“…In particular, many people worked on the problem of finding a probabilistic approximation of an arbitrary belief function. Several papers [2][3][4][5][6][7][8][9][10] have been published on this issue, mainly in order to find efficient implementations of the rule of combination aiming to reduce the number of focal elements. The connection between belief functions and probabilities is as well crucial in Smets' "Transferable Belief Model" [11].…”

Semantics of the Relative Belief of Singletons

“…Tessem [21], for instance, incorporated only the highest-valued focal elements in his m klx approximation; a similar approach inspired the summarization technique formulated by Lowrance et al [15]. In Smets' "Transferable Belief Model" [17] beliefs are represented at credal level (as convex sets of probabilities), while decisions are made by resorting to a Bayesian belief function called pignistic transformation [19].…”