rapaport@buffalo.CSNET ("William J. Rapaport") (03/13/87)
SECOND ANNUAL SUNY BUFFALO GRADUATE CONFERENCE ON COMPUTER SCIENCE William J. Rapaport Department of Computer Science SUNY Buffalo Buffalo, NY 14260 rapaport@buffalo.csnet On 10 March 1987, the graduate students in the Department of Computer Science at SUNY Buffalo held their second annual Graduate Conference on Computer Science. (For a report on the first one, see SIGART No. 99, pp. 22-24.) This time, the conference took on an international flavor, with talks by graduate students from the University of Toronto and the University of Rochester, in addition to talks by our own students. Once again, the conference was flawlessly mounted. The conference was sponsored by the SUNY Buffalo Department of Computer Science, the SUNY Buffalo Computer Science Graduate Student Association, the SUNY Buffalo Graduate Student Association, and the Niagara Frontier Chapter of the ACM. Approximately 150 people from area colleges and industry attended. A SUNY Buffalo Department of Computer Science Technical Report with extended abstracts of the talks (James Geller & Keith Bettinger (eds.), _UBGCSS-87: Proceedings of the Second Annual UB Graduate Conference on Computer Science_, Technical Report 87-04, March 1987) is available by contacting the chair of the organizing committee, Scott Campbell, Department of Computer Science, SUNY Buffalo, Buffalo, NY 14260, campbl@buffalo.csnet. Following are the abstracts of the talks. Ted F. Pawlicki SUNY Buffalo "The Representation of Visual Knowledge" This paper reports on preliminary research into the representation of knowledge necessary for visual recognition. The problem is broken down into three parts: the actual knowledge that needs to be represented, the form that the representation should take, and how the knowledge itself and its representation should combine to facilitate the visual recognition task. The knowledge chosen to represent is a formalization of the theory of Recognition by Component. The representation chosen is a semantic network. John M. Mellor-Crummey University of Rochester "Parallel Program Debugging with Partial Orders" Parallel programs are considerably more difficult to debug than sequential programs, because successive executions of a parallel program often do not exhibit the same behavior. Instant Replay is a new technique for reproducing parallel-program executions. Partial orders of significant events are recorded during program execution and used to enforce equivalence of execution replays. This technique (1) requires less time and space to save information for program replay than other methods, (2) is independent of the form of interprocess communication, (3) provides for replay of an entire program, rather than individual processes, (4) introduces no centralized bottlenecks, and (5) does not require synchronized clocks or globally-consistent logical time. Some performance results of a prototype on the BBN Butterfly [TM] Parallel Processor will be presented, and it will be shown how Instant Replay can be used in the debugging cycle for parallel programs. Timothy D. Thomas and Susan J. Wroblewski SUNY Buffalo "Efficient Trouble Shooting in an Industrial Environment" Our work involves designing and implementing a real-time system for trouble shooting in an industrial environment. The system emulates the kind of problem-solving knowledge and behavior typical of a human expert after years of on-the-job experience. Our system, PASTE (Process Analysis for Solving Trouble Efficiently), is to be used in a real-time environment. It is because of this constraint that the design of an efficient system was of great importance. PASTE has a number of efficiency techniques that eliminate redundancy in remedy suggestion and that decrease response time. Ching-Huei Wang SUNY Buffalo "ABLS: An Object Recognition System for Locating Address Blocks on Mail Pieces" ABLS (Address Block Location System), a system for locating address blocks on mail pieces, represents both a specific solution to postal automation and a general framework for coordinating a collection of specialized image-processing tools to opportunistically detect objects in images. Images that ABLS deals with range from those having a high degree of global spatial structure (e.g., carefully prepared letter mail envelopes which conform to specifications) to those with no structure (e.g., magazines with randomly pasted address labels). Its problem-solving architecture is based on the blackboard model and utilizes a dependency graph, knowledge rules, and a blackboard. Diane Horton and Graeme Hirst University of Toronto "Presuppositions as Beliefs: A New Approach" Most existing theories of presupposition implicitly assume that presuppositions are facts and that all agents involved in a discourse share belief in the presuppositions that it generates. We argue that these assumptions are unrealistic and can be eliminated by treating each presupposition as the belief of an agent. We describe a new model, including an improved definition of presupposition, that takes this approach. The new model is more realistic and can handle cases of presupposition projection that could not be handled otherwise. Norman D. Wahl and Susan E. Miller SUNY Buffalo "Hypercube Algorithms to Determine Geometric Properties of Digitized Pictures" This research focuses on implementing algorithms to solve geometric problems of digitized pictures on hypercube multiprocessors. Specifically, in this paper, we present algorithms and paradigms for solving the connected component labeling problem. Work is ongoing to complete implementations of these algorithms and obtain running times on the Intel iPSC and Ncube hypercubes. The goal of this study is to determine under what circumstances (if any) each of the various algorithms is most appropriate. Deborah Walters and Ganapathy Krishnan SUNY Buffalo "Bottom-up Image Analysis for Color Separation" A system for automatic color separation for use in the printing industry is described. The goal of this research was to automate the labor-intensive preprocessing required before a graphics system can process the image. This system makes no assumptions about the semantic content of the image. The processing is entirely bottom-up and is based on image features used by the human visual system during the early stages of processing. The image is convolved with oriented edge operators, and the responses are stored in the Rho-Space representation. A number of parallel operations are performed in Rho-Space, and the image is segmented into perceptually significant parts, which can then be colored using an interactive graphics system. Bart Selman University of Toronto "Vivid Representations and Analogues" Levesque introduced the notion of a vivid knowledge representation. A vivid scheme contains complete knowledge in a tractable form. A closely related concept is that of an analogical representation or analogue. Sloman characterizes analogues as representations that are in some sense direct models of the domain, as opposed to representations consisting of a description in some general language. The prototypical example of an analogical representation is a pictorial representation, which is also an important source of vivid knowledge. We are studying these types of representations for their possible application in computationally tractable knowledge-representation systems. In particular, we are studying how information in a non-analogical (or non-vivid) form can be translated into an analogical (or vivid) form, using for example defaults and prototypes. This talk will cover the properties of vivid and analogical representations, a description of their relationship to each other, and some initial ideas on the translation process. Soteria Svorou SUNY Buffalo "The Semantics of Spatial Extension Terms in Modern Greek" In recent years, there have been increasing efforts to uncover the nature of the human mind by studying the structure of its building blocks: concepts. Partaking in this enterprise, this study explores the domain of spatial extension categories by looking at the way language treats them. It shows that lexical contrasts of Modern Greek in the domain of spatial extension reflect the perceptual strategies of "orientation" and "Gestalt" and their interaction with the concept of "boundedness", which speakers employ in the description of everyday objects. Yong Ho Jang and Hing Kai Hung SUNY Buffalo "Semantics of a Recursive Procedure with Parameters and Aliasing" We consider a subset of an Algol-like programming language that includes blocks and recursive procedures, with value and location parameter passing. We develop the operational and denotational semantics for both static and dynamic scope, with their different aliasing mechanisms. The main advantage of our approach is that the denotational semantics is compositional and can systematically handle the various scope and aliasing features. Josh D. Tenenberg and Leo B. Hartman University of Rochester "Naive Physics and the Control of Inference" Hayes proposed the naive physics program in order eventually to address problems involving the control of inference. At the time of the proposal, progress toward solutions of these problems seemed impeded by the lack of a well-defined body of knowledge of challenging size. The building of a formally interpretable encoding of the common-sense knowledge that people use to deal with the physical world seemed to fill this need. It was argued that the knowledge be expressed in first-order logic or an equivalent language in order to separate declarative information from control information. We argue here that no finite encoding of a formal theory can be completely separated from control choices by virtue of there being well-defined measures of the depth of a theorem in the deductive closure of a theory. In addition, any control choice is a commitment to a particular set of statistical properties of the problems an agent faces, and the measurement of such properties is required to evaluate these choices. Zhigang Xiang SUNY Buffalo "Multi-Level Model-Based Diagnostic Reasoning" Diagnostic systems capable of reasoning from _functional_ and _structural_ knowledge are _model-based_ systems. The uniqueness of our work is that problems of diagnosis that need not only functional and _logical_ structural knowledge but also _spatial_ structural knowledge are to be the focus. Towards this goal, we propose a framework for organizing, representing, and reasoning with an integrated knowledge base that includes multiple levels of abstraction of the physical system. More specifically, a physical system is decomposed into physical and logical components. Analogical (geometrical) and propositional (topological) spatial structural information are associated with physical components. The latter is mutually related to logical components. Functional relationships are established between logical components. Logical reasoning infers the functional status of logical components, whereas spatial reasoning performs fault localization. The framework is carried out using semantic-network representations. The implementation is independent of any given domain of application. The system, when given a description of a physical system's spatial structure, logical structure, and functional relationships between logical components, performs logical as well as spatial reasoning to locate faulty components, lesions, etc., from symptoms and findings. Domain-specific examples include circuitry fault localization and neuroanatomic localization.