Michael H. Smith

KASER: knowledge amplification by structured expert randomization

In this paper and attached video, we present a third-generation expert system named Knowledge Amplification by Structured Expert Randomization (KASER) for which a patent has been filed by the U.S. Navys SPAWAR Systems Center, San Diego, CA (SSC SD). KASER is a creative expert system. It is capable of deductive, inductive, and mixed derivations. Its qualitative creativity is realized by using a tree-search mechanism. The system achieves creative reasoning by using a declarative representation of knowledge consisting of object trees and inheritance. KASER computes with words and phrases. It possesses a capability for metaphor-based explanations. This capability is useful in explaining its creative suggestions and serves to augment the capabilities provided by the explanation subsystems of conventional expert systems. KASER also exhibits an accelerated capability to learn. However, this capability depends on the particulars of the selected application domain. For example, application domains such as the game of chess exhibit a high degree of geometric symmetry. Conversely, application domains such as the game of craps played with two dice exhibit no predictable pattern, unless the dice are loaded. More generally, we say that domains whose informative content can be compressed to a significant degree without loss (or with relatively little loss) are symmetric. Incompressible domains are said to be asymmetric or random. The measure of symmetry plus the measure of randomness must always sum to unity.

Automatic design and tuning of a fuzzy system for controlling the Acrobot using genetic algorithms, DSFS, and meta-rule techniques

In this paper, we present practical methods for automatic design and tuning of fuzzy systems and apply them to a complex control problem: swing-up control of a two-link robot called the Acrobot. We use genetic algorithms, dynamic switching fuzzy systems (DSFS), and meta-rule techniques to obtain a high performance meta-rule enhanced TSK controller for the Acrobot. Our paper demonstrates how we integrate these techniques and how they allow us to reduce design time and system complexity.<<ETX>>

A granular signature of data

In this paper, we discuss an issue of description of highly dimensional data realized in terms of fuzzy sets. The underlying idea is to granulate numeric data using fuzzy sets and afterwards reveal and quantify relationships between these granules. This naturally impacts the dimensionality of any original dataset under discussion and provides with its nonlinear transformation (through the corresponding membership functions). These information granules give rise to the notion of associations-multidimensional information granules. Being fuzzy relations, these constructs are direction free. The directionality arises when one defines inputs and outputs and in this way confines himself to some sort of rules capturing a directional nature of main relationships within the data. Rules arising from associations may be in conflict. The essence of data is then captured via a granular signature regarded as a mixture of associations and rules.

Design limitations of PD versus fuzzy controllers for the Acrobot

This paper discusses the swing-up control of a two link robot mechanism known as the Acrobot. First a PD controller is investigated and then a fuzzy controller is implemented by a combination of genetic algorithms, dynamic switching fuzzy systems, and meta-rule techniques. Limitations of the PD controller and the fuzzy controller are shown by investigating the effect of external random disturbances on the Acrobot using simulation. Design considerations are discussed. This study illustrates how intelligent control methods can reduce development effort and system complexity while achieving better performance.

A learning program which plays partnership dominoes

A learning program has been written in BASIC to play four-player partnership dominoes. Because dominoes is a game of incomplete information, the program uses somewhat different principles of artificial intelligence from those used in programs for games of complete information, such as checkers, chess, and go. The program was constructed to use a “strategy signature table” which classifies board situations through the interactions of game parameters. Each entry in the table contains adaptively determined weights indicating the advisability of various strategies. Once chosen, a strategy then employs probability analysis and linear polynomial evaluation to choose a move. Our program wins approximately two-thirds of its games in tournament situations, and has defeated championship players.

KASER: a qualitatively fuzzy object-oriented inference engine

This paper describes a shell that has been developed for the purpose of fuzzy qualitative reasoning. The relation among object predicates is defined by object trees that are fully capable of dynamic growth and maintenance. The qualitatively fuzzy inference engine and the developed expert system can then acquire a virtual-rule space that is exponentially (subject to machine implementation constants) larger than the actual, declared-rule space and with a decreasing non-zero likelihood of error. This capability is called knowledge amplification, and the methodology is named KASER. KASER is an acronym for Knowledge Amplification by Structured Expert Randomization. It can handle the knowledge-acquisition bottleneck in expert systems. KASER represents an intelligent, creative system that fails softly, learns over a network, and has enormous potential for automated decision making. KASERs compute with words and phrases and possess capabilities for metaphorical explanations.

Granular correlation analysis in data mining

In this paper, we introduce and study fuzzy sets in the context of granular correlation. This arises as a result of using fuzzy information granules and can be regarded as a generic vehicle for data mining. It is shown how an analysis of fuzzy granular correlation helps to reveal and quantify relationships between variables in any data mining task.

Fuzzy data mining for querying and retrieval of research archival information

This paper discusses the design of a prototype information/intelligent system (FUZZYBASE) to facilitate intelligent and fast retrieval of information that is of interest to scientific research communities with specific needs, such as getting relevant technical information fast. It involves the intelligent fuzzy retrieval of information, crisp retrieval of catalogued information, high end computer and communications infrastructure, the potential to be a testbed for future research in databases, retrieval algorithms, and networking, and so on. The prototype components of our design include a subsystem to automatically build a profile for each user using their CV or other information, a push technology subsystem, a new electronic journal, tutorial links to sites with relevant scientific information and research, pointers to electronic abstracts, tables of contents of other journals as well as other WWW pages, and links to other relevant sites, links to various conferences, calls for papers, abstracts, and conference papers, a bulletin board, a querying or information retrieval system, and a front end to the fuzzy retrieval component with a specification subsystem using natural language.

Designing a fuzzy controller for the Acrobot to compensate for external disturbances

Discusses the swing-up control of a two link robot mechanism known as the Acrobot. In previous papers, a PD controller was investigated and then a better performing fuzzy controller was implemented by a combination of genetic algorithms, dynamic switching fuzzy systems, and meta-rule techniques. Limitations of the PD controller and the fuzzy controller were studied by investigating the effect of external random disturbances on the Acrobot using simulation. This paper illustrates how the fuzzy controller can be tuned to adjust for unknown external random disturbances.

Mapping the Internet

Discovery of a network topology is a challenging task. Available algorithms that rely on simple hop-limited, traceroute-style probes give different performance in terms of the completeness (fuzziness) of the resulting map, the speed of mapping, and the efficiency of mapping. The authors provide a brief overview of the types of mapping abstractions that have been used and review available techniques for generating maps of the Internets infrastructure. A small study conducted in order to compare two of these techniques is described. Results of this study indicate that informed random address probing offers more complete network maps quickly and more efficiently. They also suggest that probing from multiple sources and amalgamating the results may improve the completeness of maps.