AI THESES Received: from Mars.UCC.UMass.EDU by UMass (outbound name server) with BSMTP; 3 Mar 88 01:37:33 EST Date: Thu, 3 Mar 88 00:59 EDT From: krovetz@UMass To: e1ar0002@smuvm1.bitnet Subject: AI-Related Dissertations The following is a list of dissertation titles and abstracts related to Artificial Intelligence taken taken from the Dissertation Abstracts International (DAI) database. The list is assembled by Susanne Humphrey and myself and is published in the SIGART Newsletter (that list doesn't include the abstracts). The dissertation titles and abstracts contained here are published with the permission of University Microfilms International, publishers of the DAI database. University Microfilms has granted permission for this list to be redistributed electronically and for extracts and hardcopies to be made of it, provided that this notice is included and provided that the list is not sold. Copies of the dissertations may be obtained by addressing your request to: University Microfilms International Dissertation Copies Post Office Box 1764 Ann Arbor, Michigan 48106 or by telephoning (toll-free) 1-800-521-3042 (except for Michigan, Hawaii, and Alaska). In Canada: 1-800-268-6090. From SIGART Newsletter, No. 101 File 1 of 3 Business Admin through Comput Sci ---------------------------------------------------------------- AN University Microfilms Order Number ADG87-00409. AU LEE, JAE BEOM. IN New York University, Graduate School of Business Administration Ph.D. 1986, 228 pages. TI INTELLIGENT DECISION SUPPORT SYSTEMS FOR BUSINESS APPLICATIONS: WITH AN EXAMPLE OF PORTFOLIO MANAGEMENT DECISION MAKING. SO DAI V47(09), SecA, pp3473. DE Business Administration, General. Information Science. AB This study involves exploratory research to develop more effective approaches to designing man-machine interfaces. A key objective is to improve DSS development methodologies to allow incorporation of expert system (ES) and artificial intelligence (AI) techniques. We identify the following problems in using AI techniques for DSS development: (1) selection of the software component types (a database model, management science model, or knowledge representation and inferencing scheme) that best fit the tasks to be performed and are appropriate for the users; (2) acquisition of the appropriate predefined software components or construction of new ones; (3) combination of the heterogeneous components into a useful system. An Intelligent Decision Support System (IDSS) has been proposed to solve the above problems. A rough architecture for an IDSS has been developed that supports the business problem-solving environment by employing a useful subset of knowledge representation and inference techniques. A prototype system for portfolio management decision making has been implemented in Prolog to illustrate and validate the approach. The research makes several practical contributions in the area of DSS analysis and design. Among them are: (1) a new ES development strategy for business problem-solving environments; (2) the architecture for the IDSS which recognizes the need for multiple knowledge representation schemes; (3) the knowledge engineering techniques which can be used as guidelines for other ES developers. AN University Microfilms Order Number ADG87-03235. AU LIANG, TING-PENG. IN University of Pennsylvania Ph.D. 1986, 223 pages. TI TOWARD THE DEVELOPMENT OF A KNOWLEDGE-BASED MODEL MANAGEMENT SYSTEM. SO DAI V47(10), SecA, pp3803. DE Business Administration, General. AB Model management systems (MMS) are the most important but least researched component of decision support systems. Recently, research in MMS design has increased dramatically because significant progress in artificial intelligence, especially in the areas of knowledge representation, heuristics, and automatic reasoning, and experience gained from developing knowledge-based systems or expert systems, have provided a very good basis for developing MMSs. Successful development of an MMS has now become a promising and challenging research topic for researchers in information system areas. Because of the similarity between a data base and a model base, many previous researchers have focused on applying existing data models, such as the relational model, to the development of model management systems. However, in addition to the functions similar to data base management systems, such as model storage, retrieval, execution, and maintenance, a knowledge-based management system needs the following two capabilities: (1) Model Integration. A mechanism for integrating existing models so that the model in the model base is not only a stand-alone model but also a module for creating ad hoc models. (2) Model Selection. a mechanism that facilitates the process of model selection. This dissertation focuses on applying artificial intelligence techniques, especially the automated reasoning capabilities for model integration and selection. It first proposes a conceptual framework for MMS design which integrates four different considerations: three user's roles, three levels of models, three views of a model base, and two roles of model management systems. Secondly, a graph-based approach to model management is developed. The approach formulates the modeling process as a process for the creation of a directed network graph, which represents all candidate models for solving a problem, and the selection of a path on the network. Mechanisms and strategies for formulating a model graph are discussed. Finally, two prototypes, TIMMS (The Integrated Model Management System) and MODELCAL are presented to demonstrate the feasibility of the framework developed in this research. TIMMS is implemented in PROLOG and MODELCAL is developed in TURBO PASCAL. AN University Microfilms Order Number ADG86-29661. AU SVIOKLA, JOHN JULIUS. IN Harvard University D.B.A. 1986, 448 pages. TI PlanPower, XCON, and MUDMAN: AN IN-DEPTH ANALYSIS INTO THREE COMMERCIAL EXPERT SYSTEMS IN USE. SO DAI V47(09), SecA, pp3473. DE Business Administration, General. AB The objective of this thesis is to generate knowledge about the effects of ESs on the organizations which use them. Three field sites with expert systems in active use are examined, and the implications for management are drawn from the empirical observations. This thesis uses a comparative, three-site, pre-post exploratory design to describe and compare the effects of ES use on three organizations: The Financial Collaborative (using the PlanPower system), Digital (XCON) and Baroid (MUDMAN). The study is guided by the notions of organizational programs as defined by March and Simon, and the information-processing capacity of the firm, as defined by Galbraith, to organize, describe, and compare the effects of ES use across the sites. Eleven exploratory hypotheses act as a basis for theory-building and further hypothesis generation. ESs address ill-structured problems. Ill-structured problems are those problems for which the solution methods and criteria are either ill-defined or non-existent. In investigating three large-scale ESs in use, this researcher discovered that these systems seem to create a phenomenon referred to as "progressive structuring." This process alters the nature of the underlying task and the organizational mechanisms which support it. This phenomenon is dynamic and evolves over time. All the ESs seemed to increase the effectiveness and efficiency of the user firm. The price of the benefits was an increased rigidity in the task. In considering ESs a manager should be concerned not only with the ES itself, but with the process by which the ES is adapted, and the overall process of creating and using the ES. In addition, the manager needs to consider the effects of the ES on the uncertainty associated with the task and should consciously manage that uncertainty to foster the level of adaptation necessary to keep the ES alive and viable in the organization. AN University Microfilms Order Number ADG87-01132. AU ANDERSON, CHARLES WILLIAM. IN University of Massachusetts Ph.D. 1986, 260 pages. TI LEARNING AND PROBLEM SOLVING WITH MULTILAYER CONNECTIONIST SYSTEMS. SO DAI V47(09), SecB, pp3846. DE Computer Science. AB The difficulties of learning in multilayered networks of computational units has limited the use of connectionist systems in complex domains. This dissertation elucidates the issues of learning in a network's hidden units, and reviews methods for addressing these issues that have been developed through the years. Issues of learning in hidden units are shown to be analogous to learning issues for multilayer systems employing symbolic representations. Comparisons of a number of algorithms for learning in hidden units are made by applying them in a consistent manner to several tasks. Recently developed algorithms, including Rumelhart, et al.'s, error back-propagation algorithm and Barto, et al.'s, reinforcement-learning algorithms, learn the solutions to the tasks much more successfully than methods of the past. A novel algorithm is examined that combines aspects of reinforcement learning and a data-directed search for useful weights, and is shown to out perform reinforcement-learning algorithms. A connectionist framework for the learning of strategies is described which combines the error back-propagation algorithm for learning in hidden units with Sutton's AHC algorithm to learn evaluation functions and with a reinforcement-learning algorithm to learn search heuristics. The generality of this hybrid system is demonstrated through successful applications to a numerical, pole-balancing task and to the Tower of Hanoi puzzle. Features developed by the hidden units in solving these tasks are analyzed. Comparisons with other approaches to each task are made. AN University Microfilms Order Number ADG86-29669. AU BASU, DIPAK. IN City University of New York Ph.D. 1986, 184 pages. TI MECHANIZATION OF DATA MODEL DESIGN: A PETRI NET BASED APPROACH FOR LEARNING. SO DAI V47(09), SecB, pp3846. DE Computer Science. AB Development and design of data models plays an important role in the mechanization of solution of problems. In this dissertation, we discuss mechanization of the design of data models. We focus our attention to the micro-world of combinatorial problems, their solutions, and the data models for the solutions. We show how a model can be constructed for the micro-world. We discuss how a machine can learn to construct such a model when it is provided with a rudimentary data model consisting of rules and definitions of a problem. For this purpose, we interpret the states of the problem and the actions that connect the states, as place-nodes and transition-nodes respectively, of a Petri net: a bipartite directed multi-graph. The petri net is thought to represent the dynamics of the problem. A compatible data model based on the Petri net is constructed which supports and drives the Petri net. This enables the machine to solve the combinatorial problem at hand proving the effectiveness of the data model. We use a heirarchical learning process to enable the machine to construct the Petri net and the corresponding data model. This evolutionary approach to data model design is viewed as mechanization of design of such models. AN University Microfilms Order Number ADG87-02684. AU BELEW, RICHARD KUEHN. IN The University of Michigan Ph.D. 1986, 328 pages. TI ADAPTIVE INFORMATION RETRIEVAL: MACHINE LEARNING IN ASSOCIATIVE NETWORKS. SO DAI V47(10), SecB, pp4216. DE Computer Science. AB One interesting issue in artificial intelligence (AI) currently is the relative merits of, and relationship between, the "symbolic" and "connectionist" approaches to intelligent systems building. The performance of more traditional symbolic systems has been striking, but getting these systems to learn truly new symbols has proven difficult. Recently, some researchers have begun to explore a distinctly different type of representation, similar in some respects to the nerve nets of several decades past. In these massively parallel, connectionist models, symbols arise implicitly, through the interactions of many simple and sub-symbolic elements. One of the advantages of using such simple elements as building blocks is that several learning algorithms work quite well. The range of application for connectionist models has remained limited, however, and it has been difficult to bridge the gap between this work and standard AI. The AIR system represents a connectionist approach to the problem of free-text information retrieval (IR). Not only is this an increasingly important type of data, but it provides an excellent demonstration of the advantages of connectionist mechanisms, particularly adaptive mechanisms. AIR's goal is to build an indexing structure that will retrieve documents that are likely to be found relevant. Over time, by using users' browsing patterns as an indication of approval, AIR comes to learn what the keywords (symbols) mean so as use them to retrieve appropriate documents. AIR thus attempts to bridge the gap between connectionist learning techniques and symbolic knowledge representations. The work described was done in two phases. The first phase concentrated on mapping the IR task into a connectionist network; it is shown that IR is very amenable to this representation. The second, more central phase of the research has shown that this network can also adapt. AIR translates the browsing behaviors of its users into a feedback signal used by a Hebbian-like local learning rule to change the weights on some links. Experience with a series of alternative learning rules are reported, and the results of experiments using human subjects to evaluate the results of AIR's learning are presented. AN This item is not available from University Microfilms International ADG05-59521. AU CHAN, KWOK HUNG. IN The University of Western Ontario (Canada) Ph.D. 1986. TI FOUNDATIONS OF LOGIC PROGRAMMING WITH EQUALITY. SO DAI V47(10), SecB, pp4217. DE Computer Science. AB An obstacle to practical logic programming systems with equality is infinite computation. In the dissertation we study three strategies for eliminating infinite searches in Horn clause logic programming systems and develop an extension of Prolog that has the symmetry, transitivity and predicate substitutivity of equality built-in. The three strategies are: (1) Replacing logic programs with infinite search trees by equivalent logic programs with finite search trees; (2) Building into the inference machine the axioms that cause infinite search trees; (3) Detecting and failing searches of infinite branches. The dissertation consists of two parts. General theories of the three strategies identified above are developed in Part I. In Part II we apply these strategies to the problem of eliminating infinite loops in logic programming with equality. Part I. General Theories. We introduce the notion of CAS-equivalent logic programs: logic programs with identical correct answer substitutions. Fixpoint criteria for equivalent logic programs are suggested and their correctness is established. Semantic reduction is introduced as a means of establishing the soundness and completeness of extensions of SLD-resolution. The possibility of avoiding infinite searches by detecting infinite branches is explored. A class of SLD-derivations called repetitive SLD-derivation is distinguished. Many infinite derivations are instances of repetitive SLD-derivations. It is demonstrated that pruning repetitive SLD-derivations from SLD-trees does not cause incompleteness. Part II. Extended Unification for Equality. An extension of SLD-resolution called SLDEU-resolution is presented. The symmetry, transitivity and predicate substitutivity of equality are built into SLDEU-resolution by extended unification. Extended unification, if unrestricted, also introduces infinite loops. We can eliminate some of these infinite loops by restricting SLDEU-resolution to non-repetitive right recursive SLDEU-resolution; this forbids extended unification of the first terms in equality subgoals and has a built-in mechanism for detecting repetitive derivations. The soundness and completeness of non-repetitive right recursive SLDEU-resolution are proved. AN University Microfilms Order Number ADG87-01100. AU COOPER, NELL. IN The University of Texas at Arlington Ph.D. 1986, 117 pages. TI A FORMAL DESCRIPTION AND THEORY OF KNOWLEDGE REPRESENTATION METHODOLOGIES. SO DAI V47(09), SecB, pp3847. DE Computer Science. AB The absence of a common and consistently applied terminology in discussions of knowledge representation techniques and the lack of a unifying theory or approach are identified as significant needs in the area of knowledge representation. Knowledge representation viewed as a collection of levels is presented as an alternative to traditional definitions. The levels and their associated primitives are discussed. The concept of levels within each knowledge representation technique provides resolution to many of the controversies and disagreements that have existed among researchers concerning the equivalency of representation methodologies. A statement of the equivalence of a certain class of frame knowledge representation and a certain class of logic based knowledge representation is presented. Definitions of the classes are included. Algorithms to convert from each class to the other are given as evidence of their equivalence. AN University Microfilms Order Number ADG87-03200. AU DURRANT-WHYTE, HUGH FRANCIS. IN University of Pennsylvania Ph.D. 1986, 235 pages. TI INTEGRATION, COORDINATION AND CONTROL OF MULTI-SENSOR ROBOT SYSTEMS. SO DAI V47(10), SecB, pp4219. DE Computer Science. AB This thesis develops a theory and methodology for integrating observations from multiple disparate sensor sources. An architecture for a multi-sensor robot system is proposed, based on the idea of a coordinator guiding a group of expert sensor agents, communicating through a blackboard facility. As description of the robot environment is developed in terms of a topological network of uncertain geometric features. Techniques for manipulating, transforming and comparing these representations are described, providing a mechanism for combining disparate observations. A general model of sensor characteristics is developed that describes the dependence of sensor observations on the state of the environment, the state of the sensor itself, and other sensor observations or decisions. A constrained Bayesian decision procedure is developed to cluster and intergrate sparse, partial, uncertain observations from diverse sensor systems. Using the network network topology of the world model, a method is developed for updating uncertain geometric descriptions of the environment in a manner that maintains a consistent interpretation for observations. A team theoretic representation of dynamic sensor operation is used to consider competitive, complementary, and cooperative elements of multi-sensor coordination and control. These descriptions are used to develop algorithms for the dynamic exchange of information between sensor systems and the construction of active sensor strategies. This methodology is implemented on a distributed computer system using an active stereo camera and a robot-mounted tactile gripper. AN University Microfilms Order Number ADG87-02882. AU EBELING, WILLIAM HENRY CARL. IN Carnegie-Mellon University Ph.D. 1986, 187 pages. TI ALL THE RIGHT MOVES: A VLSI ARCHITECTURE FOR CHESS. SO DAI V47(10), SecB, pp4219. DE Computer Science. AB Hitech, the Carnegie-Mellon chess program that recently won the ACM computer chess championship and owns a USCF rating of 2352, owes its success in large part to an architecture that embraces both move generation and position evaluation. Previous programs have been subject to a tradeoff between speed and knowledge: applying more chess knowledge to position evaluation necessarily slows the search. Recent experience with chess programs such as Belle, Cray Blitz and BEBE has shown that a deep search solves many problems that a shallow search with deep understanding cannot cope with. With this new architecture, Hitech is able to search both deeply and knowledgeably. Chapter 2 gives some background and describes previous hardware move generators. This chapter discusses the requirements of the move generator in light of the performance of the (alpha)-(beta) search. Chapter 3 presents a new architecture for move generation which allows fine-grained parallelism to be applied with very effective results. Although the amount of hardware required is substantial, the architecture is eminently suited to VLSI. This chapter also gives the details of the move generator used by Hitech, which comprises 64 identical custom VLSI chips. This move generator is able to judge moves much more effectively than previous move generators because it knows all the moves available for each side. Since the efficiency of the (alpha)-(beta) search depends on the order in which moves are examined, this ability of the move generator to order moves extremely well results in a very efficient search. Chapter 4 describes the requirements of position evaluation and discusses how this architecture can be used to perform evaluation as well. This includes a description of a VLSI implementation that we propose for position evaluation. Chapter 5 describes the other Hitech hardware and software. Chapter 6 presents a performance analysis of Hitech as a whole and the move generator in particular. Based on these measurements, some ways to improve the move generator performance are discussed. Finally, we draw some conclusions about the effect of the architecture presented in this thesis on the problem of chess. AN University Microfilms Order Number ADG87-01496. AU GREENBAUM, STEVEN. IN University of Illinois at Urbana-Champaign Ph.D. 1986, 259 pages. TI INPUT TRANSFORMATIONS AND RESOLUTION IMPLEMENTATION TECHNIQUES FOR THEOREM PROVING IN FIRST-ORDER LOGIC. SO DAI V47(09), SecB, pp3848. DE Computer Science. AB This thesis describes a resolution based theorem prover designed for users with little or no knowledge of automated theorem proving. The prover is intended for high speed solution of small to moderate sized problems, usually with no user guidance. This contrasts with many provers designed to use substantial user guidance to solve hard or very hard problems, often having huge search spaces. Such provers are often weak when used without user interaction. Many of our methods should be applicable to large systems as well. Our prover uses a restricted form of locking resolution, together with an additional resolution step. Pending resolvents are ordered using a priority-based search strategy which considers a number of factors, including clause complexity measures, derivation depth of the pending resolvent, and other features. Also described are transformations that convert formulas from one to another. One is a nonstandard clause-form translation which often avoids the loss of structure and increase in size resulting from the conventional translation, and also takes advantage of repeated subexpressions. Another transformation replaces operators in first-order formulas with their first-order definitions, before translation to clause form. This works particularly well with the nonstandard clause-form translation. There is also a translation from clauses to other clauses that, when coupled with some prover extensions, is useful for theorem proving with equality. The equality method incorporates Knuth-Bendix completion into the proof process to help simplify the search. Some implementation methods are described. Data structures that allow fast clause storage and lookup, and efficient implementation of various deletion methods, are discussed. A modification of discrimination networks is described in detail. AN University Microfilms Order Number ADG87-01292. AU HARBISON-MOSS, KARAN ANN. IN The University of Texas at Arlington Ph.D. 1986, 220 pages. TI MAINTAINING CURRENT STATUS IN A TIME-CONSTRAINED KNOWLEDGE-BASED SYSTEM CHARACTERIZED BY CONTINUOUSLY INCOMING TEMPORAL DATA. SO DAI V47(09), SecB, pp3849. DE Computer Science. AB Reasoning processes for knowledge-based systems have in the past focused on maintaining a current database with a single context and a given set of data. These methods for reason maintenance do not suffice in domains in which data is acquired during the solution process and in which there is a constraint on time to decision. A reasoning process is proposed for these data acquisition time-constrained problems that allows multiple contexts and contradictions to exist. This flexibility, in turn, simplifies the retraction of data for nonmonotonic inferencing and allows direct assessment of goal state progression. This reasoning process is designed to function within the architecture of a knowledge-based system which itself was developed to meet the requirements of data acquisition time-constrained domains. AN University Microfilms Order Number ADG87-00203. AU HERMENEGILDO, MANUEL VICTOR. IN The University of Texas at Austin Ph.D. 1986, 268 pages. TI AN ABSTRACT MACHINE BASED EXECUTION MODEL FOR COMPUTER ARCHITECTURE DESIGN AND EFFICIENT IMPLEMENTATION OF LOGIC PROGRAMS IN PARALLEL. SO DAI V47(09), SecB, pp3849. DE Computer Science. AB The term "Logic Programming" refers to a variety of computer languages and execution models which are based on the traditional concept of Symbolic Logic. The expressive power of these languages offers promise to be of great assistance in facing the programming challenges of present and future symbolic processing applications in Artificial Intelligence, Knowledge-based systems, and many other areas of computing. The sequential execution speed of logic programs has been greatly improved since the advent of the first interpreters. However, higher inference speeds are still required in order to meet the demands of applications such as those contemplated for next generation computer systems. The execution of logic programs in parallel is currently considered a promising strategy for attaining such inference speeds. Logic Programming in turn appears as a suitable programming paradigm for parallel architectures because of the many opportunities for parallel execution present in the implementation of logic programs. This dissertation presents an efficient parallel execution model for logic programs. The model is described from the source language level down to an "Abstract Machine" level, suitable for direct implementation on existing parallel systems or for the design of special purpose parallel architectures. Few assumptions are made at the source language level and therefore the techniques developed and the general Abstract Machine design are applicable to a variety of logic (and also functional) languages. These techniques offer efficient solutions to several areas of parallel Logic Programming implementation previously considered problematic or a source of considerable overhead, such as the detection and handling of variable binding conflicts in AND-Parallelism, the specification of control and management of the execution tree, the treatment of distributed backtracking, and goal scheduling and memory management issues etc. A parallel Abstract Machine design is offered, specifying data areas, operation, and a suitable instruction set. This design is based on extending to a parallel environment the techniques introduced by the Warren Abstract Machine, which have already made very fast and space efficient sequential systems a reality. Therefore, the model herein presented is capable of retaining sequential execution speed similar to that of high performance sequential systems, while extracting additional gains in speed by efficiently implementing parallel execution. These claims are supported by simulations of the Abstract Machine on sample programs. AN University Microfilms Order Number ADG87-01190. AU LEE, YILLBYUNG. IN University of Massachusetts Ph.D. 1986, 150 pages. TI A NEURAL NETWORK MODEL OF FROG RETINA: A DISCRETE TIME-SPACE APPROACH. SO DAI V47(09), SecB, pp3852. DE Computer Science. AB Most computational models of the nervous systems in the past have been developed at the level of a single cell or at the level of a population of uniform elements. But neither the absolute temporal activity of a single neuron nor some steady state of a population of neurons seems to be of utmost importance. A spatio-temporal pattern of activities of neurons and the way they interact through various connections appear to matter most in the neuronal computations of the vertebrate retina. A population of neurons are modelled based on a connectionist scheme and on experimental data to provide a spatio-temporal pattern of activities for every cell involved in the cone-pathways for a patch of frog's retina. The model has discrete representations for both space and time. The density of each type/subtype of neuron and the existence and the size of the connections are based on anatomical data. Individual neurons are modelled as variations of a leaky-capacitor model of neurons. Parameters for each type of model neuron are set so that their temporal activities approximate the typical intracellular recording for the corresponding neurons given the known visual/electrical stimulus patterns. Connectivity was thought the single most important factor for network computation. Computer simulation results of the model are compared with well-known physiologic data. The results show that a network model of coarse individual neuronal models based on known structures of the vertebrate retina approximates the overall system behavior successfully reproducing the observed functions of many of the cell types, thus showing that the connectionist approach can be applied successfully to neural network modeling and provide an organizing theory of how individual neurons interact in a population of neurons. AN University Microfilms Order Number ADG87-02078. AU LEE, YONG-BOK. IN Case Western Reserve University Ph.D. 1986, 104 pages. TI CONSTRAINT PROPAGATION IN A PATHOPHYSIOLOGIC CAUSAL NETWORK. SO DAI V47(10), SecB, pp4221. DE Computer Science. AB The causal model approach to expert knowledge representation and reasoning, which is based on making the causal domain relationships explicit, is a focus of current research in expert systems. Currently existing model-based algorithms are, however, limited in the complexity of domains to which they can be applied. Recently, a semiquantitative simulation method integrated with a symbolic modeling approach based on functional and organizational primitives has been described. It has the ability to handle problems in complex domains involving nonlinear relationships between the causally related nodes. Its performance, however, requires the availability of the initial states used in the simulation. The term "initial condition" is used here to mean the specification of the values of all variables in the model at a given instant in time. These values, when then used for simulation, are "initial" in that they precede all simulated values. This thesis describes a new algorithm, called semi-quantitative inverse reasoning, for deriving a complete set of possible current state descriptions of an arbitrary complex causal model from partial specifications of the current state. Algorithms of constraint propagation by inference and hypothesis, hypothesis generation, and hypothesis conformation are developed to support the semi-quantitative inverse reasoning technique. Therefore in application to the medical domain, this technique can derive a complete set of primary diagnoses given medical data and an appropriate causal model. AN University Microfilms Order Number ADG87-00230. AU LEI, CHIN-LAUNG. IN The University of Texas at Austin Ph.D. 1986, 171 pages. TI TEMPORAL LOGICS FOR REASONING UNDER FAIRNESS ASSUMPTIONS. SO DAI V47(09), SecB, pp3852. DE Computer Science. AB In this dissertation, we consider the problem of whether the branching time or linear time framework is more appropriate for reasoning about concurrent programs in light of the criteria of expressiveness and complexity. We pay special attention to the problem of temporal reasoning under (various) fairness assumptions. In particular, we focus on the following: (1) The Model Checking Problem--Given a formula p and a finite structure M, does M define a model of p? (2) The Satisfiability Problem--Given a formula p, does there exist a structure M which defines a model of p? Algorithms for the model checking problem are useful in mechanical verification of finite state concurrent systems. Algorithms for testing satisfiability have applications not only to the automation of verification of (possibly infinite state) concurrent programs, but also in mechanical synthesis of concurrent programs where the decision procedure is used to construct a model of the specification formula from which a concurrent program is extracted. AN University Microfilms Order Number ADG86-20896. AU LI, ZE-NIAN. IN The University of Wisconsin - Madison Ph.D. 1986, 200 pages. TI PYRAMID VISION: USING KEY FEATURES AND EVIDENTIAL REASONING. SO DAI V47(09), SecB, pp3852. DE Computer Science. AB Pyramid programs and multicomputers appear to offer a number of interesting possibilities for computer visual perception. This thesis takes a pyramidal approach for the analysis of the images of cells and outdoor scenes. Transforms that compute relatively local brain-like functions are used to extract and combine features at the successive layers in the pyramid. They are applied in a parallel and hierarchical manner to model the living visual systems. The use of 'key features' in this thesis is an exploitation of the generation and use of 'focus of attention' techniques for visual perception in a pyramid vision system. In contrast to many other systems, key features are used as the central threads for the control process. They embed naturally into the pyramid structure, organizing bottom-up, top-down and lateral searches and transformations into a well-integrated structure of processes. Moveover, they are also incorporated into the knowledge representation and reasoning techniques proposed for the pyramid vision system. The term 'evidential reasoning' refers to the reasoning process conducted by a system on the basis of uncertain and incomplete data and world knowledge. The Dempster-Shafer Theory of Evidence is adapted for evidential reasoning in a multi-level pyramid vision system where images are analyzed and recognized using micro-modular production-like transforms. Treated as a belief function, the form of the system's knowledge is compact. The reasoning mechanism extends the use of the belief function and the Dempster Combination Rule. While other approaches leave a gap between the feature space and the object space, the present mapping between these two spaces makes smooth transitions. The new knowledge representation technique and its reasoning mechanism take advantage of the set-theoretic formalism, while still maintaining modularity and flexibility. The comparison between the evidential reasoning approach and a simple weight combination method shows that this new approach makes better use of the world knowledge, and offers a good way to use key features. The pyramid vision programs using key features and evidential reasoning were used successfully on two biomedical images and four outdoor-scene images. The results indicate that this new approach is efficient and effective for the analysis of complex real-world images. AN This item is not available from University Microfilms International ADG05-59587. AU LU, SIWEI. IN University of Waterloo (Canada) Ph.D. 1986. TI ATTRIBUTED HYPERGRAPH REPRESENTATION AND RECOGNITION OF 3-D OBJECTS FOR COMPUTER VISION. SO DAI V47(10), SecB, pp4221. DE Computer Science. AB This thesis presents a robot vision system which is capable of recognizing objects in a 3-D scene and interpreting their spatial relation even though some objects in the scene may be partially occluded by other objects. In my system, range data for a collection of 3-D objects placed in proximity is acquired by laser scanner. A new algorithm is developed to transform the geometric information from the range data into an attributed hypergraph representation (AHR). The AHR is a unique representation of 3-D object which is invariant to orientation. A hypergraph monomorphism algorithm is used to compare the AHR of objects in the scene with the complete AHR of a set of prototypes in a database. Through a hypergraph monomorphism, it is possible to recognize any view of an object and also classify the scanned objects into classes which consist of similar shapes. The system can acquire representation for unknown objects. Several AHR's of the various views of an unknown object can be synthesized into a complete AHR of the object which can then be included in the model database. A scene interpretation algorithm is developed to locate and recognize objects in the scene even though some of them are partially occluded. The system is implemented in PASCAL on a VAX11/750 running VMS, and the image results are displayed on a Grinnell 270 display device. AN University Microfilms Order Number ADG87-01195. AU LYONS, DAMIAN MARTIN. IN University of Massachusetts Ph.D. 1986, 255 pages. TI RS: A FORMAL MODEL OF DISTRIBUTED COMPUTATION FOR SENSORY-BASED ROBOT CONTROL. SO DAI V47(09), SecB, pp3853. DE Computer Science. AB Robot systems are becoming more and more complex, both in terms of available degrees of freedom and in terms of sensors. It is no longer possible to continue to regard robots as peripheral devices of a computer system, and to program them by adapting general-purpose programming languages. This dissertation analyzes the inherent computing characteristics of the robot programming domain, and formally constructs an appropriate model of computation. The programming of a dextrous robot hand is the example domain for the development of the model. This model, called RS, is a model of distributed computation: The basic mode of computation is the interaction of concurrent computing agents. A schema in RS describes a class of computing agents. Schemas are instantiated to produce computing agents, called SIs, which can communicate with each other via input and output ports. A network of SIs can be grouped atomically together in an Assemblage, and appears externally identical to a single SI. The senory and motor interface to RS is a set of primitive, predefined schemas. These can be grouped arbitrarily with built-in knowledge in assemblages to form task-specific object models. A special kind of assemblage called a task-unit is used to structure the way robot programs are built. The formal semantics of RS is automata theoretic; the semantics of an SI is a mathematical object, a Port Automaton. Communication, port connections, and assemblage formation are among the RS concepts whose semantics can be expressed formally and precisely. A temporal logic specification and verification method is constructed using the automata semantics as a model. While the automata semantics allows the analysis of the model of computation, the temporal logic method allows the top-down synthesis of programs in the model. A computer implementation of the RS model has been constructed, and used in conjunction with a graphic robot simulation, to formulate and test dextrous hand control programs. In general, RS facilitates the formulation and verification of versatile robot programs, and is an ideal tool with which to introduce AI constructs to the robot domain. AN University Microfilms Order Number ADG87-04023. AU MEREDITH, MARSHA JEAN EKSTROM. IN Indiana University Ph.D. 1986, 205 pages. TI SEEK-WHENCE: A MODEL OF PATTERN PERCEPTION. SO DAI V47(11), SecB, pp4584. DE Computer Science. AB Seek-Whence is an inductive learning program that serves as a model of a new approach to the programming of "intelligent" systems. This approach is characterized by: (1) structural representation of concepts; (2) the ability to reformulate concepts into new, related concepts; (3) a probabilistic, biologically-inspired approach to processing; (4) levels of abstraction in both representation and processing. The program's goals are to discover patterns, describe them as structural pattern concepts, and reformulate those concepts, when appropriate. The system should model human performance as closely as possible, especially in the sense of generating plausible descriptions and ignoring implausible ones. Description development should be strongly data-driven. Small, special-purpose tasks working at different levels of abstraction with no overseeing agent to impose an ordering eventually guide the system toward a correct and concise pattern description. The chosen domain is that of non-mathematically-sophisticated patterns expressed as sequences of nonnegative integers. A user presents a patterned number sequence to the system, one term at a time. Seek-Whence then either ventures a guess at the pattern, quits, or asks for another term. Should the system guess a pattern structure different from the one the user has in mind, the system will attempt to reformulate its faulty perception. Processing occurs in two stages. An initial formulation must first evolve; this is the work of stage one, culminating in the creation of a hypothesis for the sequence pattern. During stage two, the hypothesis is either verified or refuted by new evidence. Consistent verification will tend to confirm the hypothesis, and the system will present the user with its hypothesis. An incorrect guess or refutation of the hypothesis by new evidence will cause the system to reformulate or abandon the hypothesis. Reformulation of the hypothesis causes related changes throughout the several levels of Seek-Whence structures. These changes can in turn cause the noticing of new perceptions about the sequence, creating an important interplay among the processing levels. AN University Microfilms Order Number ADG87-00791. AU MITCHELL, JOSEPH S. B. IN Stanford University Ph.D. 1986, 143 pages. TI PLANNING SHORTEST PATHS. SO DAI V47(09), SecB, pp3853. DE Computer Science. AB Recent research in the algorithmic aspects of robot motion and terrain navigation has resulted in a number of interesting variants of the shortest path problem. A problem that arises when planning shortest collision-free paths for a robot is the following: Find the shortest path from START to GOAL for a point moving in two or three dimensions and avoiding a given set of polyhedral obstacles. In this thesis we survey some of the techniques used and some of our recent results in shortest path planning. We introduce a useful generalization of the shortest path problem, the "weighted region problem". We describe a polynomial-time algorithm which finds a shortest path through "weighted" polygonal regions, that is, which minimizes the sum of path lengths multiplied by the respective weight factors of the regions through which the path passes. Our algorithm exploits the fact that optimal paths obey Snell's Law of Refraction when passing through region boundaries. We also give an O(n('2) log n) algorithm for the special case of the three-dimensional shortest path problem in which paths are constrained to lie on the surface of a given (possibly non-convex) polyhedron. Both algorithms make use of a new technique of solving shortest path problems; we call this technique a "continuous Dijkstra algorithm", as it closely resembles the method used by Dijkstra to solve simple shortest path problems in a graph. AN University Microfilms Order Number ADG86-29095. AU MORGADO, ERNESTO JOSE MARQUES. IN State University of New York at Buffalo Ph.D. 1986, 234 pages. TI SEMANTIC NETWORKS AS ABSTRACT DATA TYPES. SO DAI V47(11), SecB, pp4584. DE Computer Science. AB Abstraction has often been used to permit one to concentrate on the relevant attributes of the domain and to disregard the irrelevant ones. This is accompanied by a reduction in the complexity of the domain. Researchers have extensively studied the use of abstraction in programming languages to allow programmers to develop software that is precise, reliable, readable, and maintainable. In spite of the amount of research that it has been subjected to, data abstraction has been largely neglected by programmers, when compared with other abstract methodologies used in programming. One problem is that it is not always easy to characterize the correct set of operations that defines an abstract data type; and, although many definitions have been presented, no precise methodology has ever been proposed to hint at the choice of those operations. A second problem is that there is a discrepancy between the formalism used to define an abstract specification and the architecture of the underlying virtual machine used to implement it. This discrepancy makes it difficult for the programmer to map the abstract specification, written at design time, into a concrete implementation, written at coding time. In order to correct these problems, a theory of data abstraction is presented, which includes a new definition of abstract data type and a methodology to create abstract data types. Because of their complexity, semantic networks are defined in terms of a variety of interrelated data types. The preciseness of the abstract data type formalism, and its emphasis on the behavior of the data type operations, rather than on the structure of its objects, makes the semantics of semantic networks clearer. In addition, the design, development, and maintenance of a semantic network processing system requires an appropriate software engineering environment. The methodology of data abstraction, with its philosophy of modularity and independence of representations, provides this kind of environment. On the other hand, the definition of a semantic network as an abstract data type and its implementation using the methodology of data abstraction provide insights on the development of a new theory of abstract data types and the opportunity for testing and refining that theory. (Abstract shortened with permission of author.). AN University Microfilms Order Number ADG87-05476. AU PEPER, GERRI L. IN Colorado State University Ph.D. 1986, 174 pages. TI INEXACT REASONING IN AN INDUCTIVE LEARNING ENVIRONMENT. SO DAI V47(11), SecB, pp4585. DE Computer Science. AB For large expert systems it is well known that better methods for acquiring expert decision-making knowledge are needed to speed up the development cycle. For this reason, there has been significant interest shown in the possibilities of using an inductive learning approach to ease this knowledge acquisition bottleneck. Although quite successful in their ability to generate correct and efficient rules, the initial attempts at inductive learning systems have failed to take into consideration a very important aspect of expert systems, that being the ability to accept and reason with uncertain knowledge. This is known as the inexact reasoning problem. This thesis describes an approach to inexact reasoning which is designed for an expert system environment which allows inductive learning as one method of knowledge acquisition. The system presented in this thesis is KNET, a generalized expert system shell which provides full support for both knowledge acquisition and consultation. It allows knowledge to be expressed in two forms, either as a set of examples or as a decision network. Transformations are allowed from one form to another. Previously existing methods of inexact reasoning have not directly dealt with these forms of knowledge representation. Three phases of the inexact reasoning problem are addressed: obtaining probabilistic knowledge during the creation of the knowledge base; using and preserving this knowledge during transformations from one form of knowledge to another; and reasoning with the inexact knowledge during the consultation. A general approach for dealing with inexact knowledge in each of these phases is presented. In addition to presenting this general approach to inexact reasoning, special consideration is given to the problem of representing uncertainty during the consultation. Emphasis is placed on insuring that the degree of uncertainty reflected by the user's answers is also clearly reflected in the certainty assigned to each of the possible conclusions presented by the system. Several possible techniques for accomplishing this task are explored. These are presented as two different models for reasoning with uncertainty. AN University Microfilms Order Number ADG87-00810. AU RENNELS, GLENN DOUGLAS. IN Stanford University Ph.D. 1986, 259 pages. TI A COMPUTATIONAL MODEL OF REASONING FROM THE CLINICAL LITERATURE. SO DAI V47(09), SecB, pp3854. DE Computer Science. AB This dissertation explores the premise that a formalized representation of empirical studies can play a central role in computer-based decision support. The specific motivations underlying this research include the following propositions: (1) Reasoning from experimental evidence contained in the clinical literature is central to the decisions physicians make in patient care. Previous researchers in medical artificial intelligence, concentrating on issues such as causal modeling, have not adequately addressed the role of experimental evidence in medical reasoning. (2) A computational model, based upon a declarative representation for published reports of clinical studies, can drive a computer program that selectively tailors knowledge of the clinical literature as it is applied to a particular case. (3) The development of such a computational model is an important first step toward filling a void in computer-based decision support systems. Furthermore, the model may help us better understand the general principles of reasoning from experimental evidence both in medicine and other domains. Roundsman is a developmental computer system which draws upon structured representations of the clinical literature in order to critique plans for the management of primary breast cancer. A distance metric has been developed to help assess the relevance of a published study to a particular clinical decision. A general model of choice and explanation in medical management has also been adapted for application to this task domain. Roundsman is able to produce patient-specific analyses of breast cancer management options based on the 24 clinical studies currently encoded in its knowledge base. Medicine will repeatedly present problem domains for which there are no reliable causal models, and in which reasoning from experimental evidence may be pivotal to problem-solving. The Roundsman system is a first step in exploring how the computer can help to bring a critical analysis of the relevant literature to the physician, structured around a particular patient and treatment decision. AN University Microfilms Order Number ADG87-05480. AU RICHARDSON, RAY CHARLES. IN Colorado State University Ph.D. 1986, 238 pages. TI INTELLIGENT COMPUTER AIDED INSTRUCTION IN STATICS. SO DAI V47(11), SecB, pp4586. DE Computer Science. AB The increased emphasis on fifth-generation computers has prompted much attention on the study of artificial intelligence and the sub-field of expert systems. Expert systems are computer programs which solve expert problems using expert knowledge. The primary emphasis of these programs is human knowledge representation of problems that humans solve. One of the areas where expert systems have been used is in education. The linking of expert systems and traditional Computer Aided Instruction is known as Intelligent Computer Aided Instruction. The purpose of this study is to demonstrate the feasibility of an expert system applied to undergraduate instruction. An expert system was developed to model the problem solving knowledge of Dr. J. L. Meriam from his text Engineering Mechanics Volume I, Statics. The rules and heuristics for solving two-dimensional truss problems were then implemented in the MRS language. The expert system was then validated by solving problems from the text in the same manner as Meriam. Linked to the expert system were three learning style modules. The learning styles modeled in this study were drill-and-practice, learning-by-example, and a new style called buddy-study. The buddy-state learning style represents an implementation of the Whimbley-pairs technique for computer based learning. The learning system comprising the expert system, learning style modules, and associated support programs were then tested for correctness and completeness. The results of the expert system validation demonstrated a system capable of solving problems within the domain as Meriam did. The learning style module testing showed procedures commensurate with accepted classroom uses of the styles. The buddy-study method demonstrated a computer learning strategy which places the expert system and the student user as colleagues in a problem solving environment. The results of the testing indicate the feasibility of such a system for inclusion in undergraduate statics courses. AN University Microfilms Order Number ADG86-25748. AU TSAO, THOMAS T. IN University of Maryland Ph.D. 1985, 151 pages. TI THE DESIGN AND ANALYSIS OF PARALLEL ADAPTIVE ALGORITHMS FOR COMPOSITE DECISION PROCESSES. SO DAI V47(09), SecB, pp3857. DE Computer Science. AB This dissertation presents new approaches to the design and analysis of parallel adaptive algorithms for multiple instruction stream multiple data stream (MIMD) machines. A composite decision process (cdp) is a model for many problems from the field of artificial intelligence. The mathematical structure modeled by a cdp includes both the algebraic structure of the domain set of the problem and the functional structure of the problem. A dynamic algorithm is a parallel algorithm with its control structure consisting of (1) an adaptive task structure, and (2) eager computation enabling mechanism. The eager computation is an enabling mechanism of parallel computations governed by processor availabilities. In the algorithm analysis, we also focus on the utility of computing power in the designed algorithm and the utility of the accumulated information in reducing the cost of search effort. The relation of these two aspects to the speed-up ratio is investigated. We call the analysis a dynamic analysis because it focuses on these major dynamic features of the parallel processes. A survey of the literature shows that very little previous work is available along these lines. The quantitative analysis presented in this dissertation confirms that in parallel adaptive search, to increase parallelism we must accept dynamic task assignment, and must have dynamic modification of global tasks in order to make best use of accumulated information. The key contributions of this dissertation to the state of the art of parallel search in A.I. are: (1) A new approach to the use of accumulated information in parallel search, whereby information accumulated during the execution of processes is used to change the specification of tasks which are remaining or unfinished. (2) A new approach to dynamic parallel algorithm design, which combines the use of accumulated (task related) information with eager computation, so that information developed during search may be employed to achieve maximum possible parallelism in a given environment. (3) A new and precise measure for speed-up obtained by a parallel algorithm. (4) A new approach to the comparative analysis of parallel algorithms. (Abstract shortened with permission of author.) AN University Microfilms Order Number ADG87-01639. AU TUCERYAN, MIHRAN. IN University of Illinois at Urbana-Champaign Ph.D. 1986, 171 pages. TI EXTRACTION OF PERCEPTUAL STRUCTURE IN DOT PATTERNS. SO DAI V47(10), SecB, pp4224. DE Computer Science. AB Perceptual grouping is an important mechanism of early visual processing. This thesis presents a computational approach to perceptual grouping in dot patterns. Detection of perceptual organization is done in two steps. The first step, called the lowest level grouping, extracts the perceptual segments of dots that group together because of their relative locations. The grouping is accomplished by interpreting dots as belonging to interior or border of a perceptual segment, or being along a perceived curve, or being isolated. The Voronoi neighborhood of a dot is used to represent its local geometric environment. The grouping is seeded by assigning to dots their locally evident perceptual roles and iteratively modifying the initial estimates to enforce global Gestalt constraints. This is done through independent modules that possess narrow expertise for recognition of typical interior dots, border dots, curve dots and isolated dots, from the properties of the Voronoi neighborhoods. The results of the modules are allowed to influence and change each other so as to result in perceptual components that satisfy global, Gestalt criteria such as border or curve smoothness and component compactness. Thus, an integration is performed of multiple constraints, active at different perceptual levels and having different scopes in the dot pattern, to infer the lowest level perceptual structure. The result of the lowest level grouping phase is the partitioning of a dot pattern into different perceptual segments or tokens. The second step further groups the lowest level tokens to identify any hierarchical structure present. The grouping among tokens is done based on a variety of constraints including their proximity, orientations, sizes, and terminations, integrated so as to mimic the perceptual roles of these criteria. This results in a new set of larger tokens. The hierarchical grouping process repeats until no new groupings are formed. The final result of the implementation described here is a hierarchical representation of the perceptual structure in a dot pattern. Our representation of perceptual structure allows for "focus of attention" through the presence of multiple levels, and for "rivalry" of groupings at a given level through the probabilistic interpretation of groupings present. AN University Microfilms Order Number ADG87-01283. AU YODER, CORNELIA MARIE. IN Syracuse University Ph.D. 1986, 383 pages. TI AN EXPERT SYSTEM FOR PROVIDING ON-LINE INFORMATION BASED ON KNOWLEDGE OF INDIVIDUAL USER CHARACTERISTICS. SO DAI V47(09), SecB, pp3858. DE Computer Science. AB In many interactive systems which provide information, such as HELP systems, the form and content of the information presented always seems to satisfy some people and frustrate others. Human Factors textbooks and manuals for interactive systems focus on the need for consistency and adherence to some standard. This implicitly assumes that if the optimum format and level of detail could be found for presenting information to a user, interactive systems would only need to adhere to the standard to be optimum for everyone. This approach neglects one of the most important factors of all--differences in people. If these individualizing differences in people could be identified, a system could be designed with options built into it to accommodate different users. The role of the intelligent active system should be more like that of a human expert or consultant, who answers questions by first interpreting them in terms of the user's knowledge and the context of his activities and then recommending actions which may be otherwise unknown to the user. The HELP system developed in this study is an Expert System written in PROLOG which uses logic programming rules to intelligently provide needed information to a terminal user. It responds to a request with a full screen display containing information determined by the request, the user's cognitive style and the user's experience level. The investigation studies the relationship between some cognitive style and experience level parameters and individual preferences and efficacy with an interactive computer information system. These factors are measured by the ability of an individual user to perform unfamiliar tasks using a HELP function as information source. The format of the information provided by the HELP function is varied along three dimensions and the content of the information is varied by three levels of detail. Experiments were performed with the system and experimental results are presented which show some trends relating cognitive style and individual preferences and performance using the system. In addition, it is argued that an Expert System can perform such a function effectively. AN University Microfilms Order Number ADG87-03940. AU YOSHII, RIKA. IN University of California, Irvine Ph.D. 1986, 152 pages. TI JETR: A ROBUST MACHINE TRANSLATION SYSTEM. SO DAI V47(11), SecB, pp4586. DE Computer Science. AB This dissertation presents an expectation-based approach to Japanese-to-English translation which deals with grammatical as well as ungrammatical sentences and preserves the pragmatic, semantic and syntactic information contained in the source text. The approach is demonstrated by the JETR system, which is composed of the particle-driven analyzer, the simultaneous generator and the context analyzer. The particle-driven analyzer uses the forward expectation-refinement process to handle ungrammatical sentences in an elegant and efficient manner without relying on the presence of particles and verbs in the source text. To achieve extensibility and flexibility, ideas such as the detachment of control structure from the word level, and the combination of top-down and bottom-up processing have been incorporated. The simultaneous generator preserves the syntactic style of the source text without carrying syntactic information in the internal representation of the text. No source-language parse tree needs to be constructed for the generator. The context analyzer is able to provide contextual information to the other two components without fully understanding the text. JETR operates without pre-editing and post-editing, and without interacting with the user except in special cases involving unknown words.