ion Modulation In many cases, it may be more useful to do normative inference on a model that is deemed to be complete at a particular level of abstraction than it is to do an approximate or heuristic analysis of a model that is too large to be analyzed under specific resource constraints. It may prove useful in many cases to store several belief-network representations, each containing propositions at different levels of abstraction. In many domains, models at higher levels of abstraction are more tractable. As the time available for computation decreases, network modules of increasing abstraction can be employed. ffl Local Reformulation Local reformulation is the modification of specific troublesome topologies in a belief network. Approximation methods and heuristics designed to modify the microstructure of belief networks will undoubtedly be useful in the tractable solution of large uncertain-reasoning problems. Such strategies might be best applied at knowledge-encoding time. An...

For agents to function effectively in large and open networks, they must ensure that their correspondents, i.e., the agents they interact with, are trustworthy. Since no central authorities may exist, the only way agents can find trustworthy correspondents is by collaborating with others to identify those whose past behavior has been untrustworthy. In other words, finding trustworthy correspondents reduces to the problem of distributed reputation management. Our approach adapts the mathematical theory of evidence to represent and propagate the ratings that agents give to their correspondents. When evaluating the trustworthiness of a correspondent, an agent combines its local evidence (based on direct prior interactions with the correspondent) with the testimonies of other agents regarding the same correspondent. We experimentally studied this approach to establish that some important properties of trust are captured by it.

Despite their different perspectives, artificial intelligence (AI) and the disciplines of decision science have common roots and strive for similar goals. This paper surveys the potential for addressing problems in representation, inference, knowledge engineering, and explanation within the decision-theoretic framework. Recent analyses of the restrictions of several traditional AI reasoning techniques, coupled with the development of more tractable and expressive decisiontheoretic representation and inference strategies, have stimulated renewed interest in decision theory and decision analysis. We describe early experience with simple probabilistic schemes for automated reasoning, review the dominant expert-system paradigm, and survey some recent research at the crossroads of AI and decision science. In particular, we present the belief network and influence diagram representations. Finally, we discuss issues that have not been studied in detail within the expert-systems sett...

Pathfinder is an expert system that assists surgical pathologists with the diagnosis of lymph-node diseases. The program is one of a growing number of normative expert systems that use probability and decision theory to acquire, represent, manipulate, and explain uncertain medical knowledge. In this article, we describe Pathfinder and our research in uncertain-reasoning paradigms that was stimulated by the development of the program. We discuss limitations with early decision-theoretic methods for reasoning under uncertainty and our initial attempts to use non-decision-theoretic methods. Then, we describe experimental and theoretical results that directed us to return to reasoning methods based in probability and decision theory.

: Belief functions are mathematical objects defined to satisfy three axioms that look somewhat similar to the Kolmogorov axioms defining probability functions. We argue that there are (at least) two useful and quite different ways of understanding belief functions. The first is as a generalized probability function (which technically corresponds to the inner measure induced by a probability function). The second is as a way of representing evidence. Evidence, in turn, can be understood as a mapping from probability functions to probability functions. It makes sense to think of updating a belief if we think of it as a generalized probability. On the other hand, it makes sense to combine two beliefs (using, say, Dempster's rule of combination) only if we think of the belief functions as representing evidence. Many previous papers have pointed out problems with the belief function approach; the claim of this paper is that these problems can be explained as a consequence of confounding the...

Most probabilistic inference algorithms are specified and processed on a propositional level. In the last decade, many proposals for algorithms accepting first-order specifications have been presented, but in the inference stage they still operate on a mostly propositional representation level. [Poole, 2003] presented a method to perform inference directly on the first-order level, but this method is limited to special cases. In this paper we present the first exact inference algorithm that operates directly on a first-order level, and that can be applied to any first-order model (specified in a language that generalizes undirected graphical models). Our experiments show superior performance in comparison with propositional exact inference. 1

We describe an ongoing project to develop technology that will support collaborative construction and effective use of distributed large-scale repositories of highly expressive reusable ontologies. We are focusing on developing a distributed server architecture for ontology construction and use, representation formalisms that remove key barriers to expressing essential knowledge in and about ontologies, ontology construction tools, and tools for obtaining domain models for use in applications from large-scale ontology repositories. We are building on the results of the DARPA Knowledge Sharing Effort, specifically by using the Knowledge Interchange Format (KIF) as a core representation language and the Ontolingua system as a core ontology development environment. In order to enable distributed ontology repositories and services, we are developing a distributed server architecture for ontology construction and use based on ontology servers which provide access via a network API to the ...

: This report presents a scene interpretation system in a context of multisensor fusion; it has been applied to the interpretation of remotely sensed images. First we present a typology of the multisensor fusion concepts involved, and we derive the consequences of modeling problems for objects, scene and strategy. The proposed multi-specialist architecture generalizes the ideas of our previous work [GG90a] by taking into account the knowledge about sensors, the multiple viewing notion (shot), and the uncertainty and impresision of models and data modeled with the Possibility Theory. In particular, generic models of objects are represented by concepts independent of sensors (geometry, materials, and spatial context). Three kinds of specialists are present in the architecture: generic specialists (scene and conflict), semantic object specialists, and low level specialists. A blackboard structure with a centralized control is used. The interpreted scene is implemented as a matrix of point...

The task of inferring a set of classes and class descriptions most likely to explain a given data set can be placed on a firm theoretical foundation using Bayesian statistics. Within this framework, and using various mathematical and algorithmic approximations, the AutoClass system searches for the most probable classifications, automatically choosing the number of classes and complexity of class descriptions. A simpler version of AutoClass has been applied to many large real data sets, have discovered new independently-verified phenomena, and have been released as a robust software package. Recent extensions allow attributes to be selectively correlated within particular classes, and allow classes to inherit, or share, model parameters though a class hierarchy. In this paper we summarize the mathematical foundations of Autoclass. 1 Introduction The task of supervised classification - i.e., learning to predict class memberships of test cases given labeled training cases - is a familia...

The theory of belief networks offers a relatively new approach for dealing with uncertain information in knowledge-based (expert) systems. In contrast with the heuristic techniques for reasoning with uncertainty employed in many rule-based expert systems, the theory of belief networks is mathematically sound, based on techniques from probability theory. It therefore seems attractive to convert existing rule-based expert systems into belief networks. In this article, we discuss the design of a belief network reformulation of the diagnostic rule-based expert system HEPAR. For the purpose of this experiment, we have studied several typical pieces of medical knowledge represented in the HEPAR system. It turned out that, due to the differences in the type of knowledge represented and in the formalism used to represent uncertainty, much of the medical knowledge required for building the belief network concerned could not be extracted from HEPAR. As a consequence, significant additional knowledge acquisition was required. However, the objects and attributes defined in the HEPAR system, as well as the conditions in production rules mentioning these objects and attributes were useful for guiding the selection of the statistical variables for building the belief network. The mapping of objects and attributes in HEPAR to statistical variables is discussed in detail.