Volume 14, 1995, No. 3


Default Reasoning in a Terminological Logic

F. Sebastiani, U. Straccia

Abstract. Terminological Logics (TLs) are knowledge representation formalisms of considerable applicative interest, as they are specifically oriented to the vast class of application domains that are describable by means of taxonomic organizations of complex objects. Although the field of TLs has lately been an active area of investigation, only few researchers have addressed the problem of extending these logics with the ability to perform default reasoning. Such extensions would prove of paramount applicative value, as many application domains may be formalized by means of monotonic TLs only at the price of oversimplification. In this paper we show how we can effectively integrate terminological reasoning and default reasoning, yielding a terminological default logic. The kind of default reasoning we embed in our TL is reminiscent of Reiter's Default Logic, but overcomes some of its drawbacks by subscribing to the "implicit" handling of exceptions typical of the Multiple Inheritance Networks with Exceptions proposed by Touretzky and others.

 

Models and Properties of Abstract Symbol Systems

D. Jakuš

Abstract. In this article we discuss properties of the abstract symbol system - a computational device proposed in [3]. We define two formal models of the abstract symbol system based on classical computational devices and study their complexities. We prove the properties of the class PP that is suggested to be an invariant class for our device. Our models are shown to be superior (in terms of the computational complexity) to the devices they originated from.  We deal with properties of complexity measures introduced for the abstract symbol system, namely computational and descriptional complexities. We prove the possibility of improvement in computational complexity for any function bounding the descriptional complexity.

 

On Epistasis

H. van Hove, A. Verschoren

Abstract. In this note, we compare two previously introduced notions  of epistasis and prove that they are essentially equivalent.

 

Armstrong Relations, Functional Dependencies and Strong Dependencies

J. Demetrovics, Vu Duc Thi

Abstract. The main purpose of this paper is to give some results related to Armstrong relations for functional dependency (FD) and strong dependency (SD). In this paper, the dependency inference problem, the FD-relation implication problem and the FD-relation equivalence problem are solved by combinatorial algorithms. We prove that the time complexity of finding a set of antikeys for a given relation scheme S is exponential in the number of attributes. Keys play important roles for the logical and structural investigation of functional dependency in the relational datamodel. We present a computational connection between relations, relation schemes, sets of antikeys and sets of minimal keys. Constructions of relation R, relation scheme S, for which KR = KS, where  KR, KS are the sets of minimal keys R, S of the FD-relation key-implication problem and the FD-relation key-equivalence problem are also presented. We show that the time complexity of finding a relation R of a given relation scheme  S such that KR = KS  is exponential in the size of S. We give a class of relations and relation schemes, for which the above problems are solved in a polynomial time. For a given relation R and a relation scheme S we introduce the following two problems:

1.    Decide whether each key of S is key of R.

2.    Decide whether all keys of R are keys of S.

We show that the first problem is solved in polynomial time, but the second problem is co-NP-complete. In the second part of the paper the concept of strong scheme is introduced. We prove that the membership problem for strong dependencies is solved by an algorithm in polynomial time. We give a necessary and sufficient condition for a relation to be Armstrong relation of a given strong scheme. We present four important problems for logical and structural investigation of strong dependencies: the construction of Armstrong relation of a given strong scheme, the construction of strong scheme SDs of which hold in a given relation, the SD-relation implication problem, the SD-relation equivalence problem.  We prove that above problems are solved by polynomial time algorithm.

 

Implementation Analysis of Fast Matrix Multiplication Algorithms on Shared Memory Computers

E. Francomano, A. Tortotici-Macaluso, M. Vajteršic

Abstract. The paper presents analysis of matrix multiplication algorithms from the point of view of their efficient implementation on shared memory machines. The algorithms of Winograd and Strassen have been analyzed and implemented considering the memory accesses, the stride and data transfer overheads. A particular attention is paid to the parallel implementation of the modified recurrence-free variant of the Strassen's algorithm.


Go To Contents