Volume 17, 1998, No. 6


knowledge interpolation: A simple approach to rapid symbolic reasoning

N. Chatterjee, J.A. Campbell

Abstract. The non-algorithmic nature of traditional knowledge-based (KB) reasoning approaches makes these techniques open-ended. As a result, KB techniques appear unsuitable for real-time applications as is evident from the small number of appearances of KB techniques in real-time domains. The technique of knowledge interpolation is a potential remedy for the shortcoming. Intuitively, the technique imitates the numerical-analysis technique of interpolation, derives solutions for an unknown problem from some already known values, and thereby avoids extensive searches of the knowledge base.  an extra inherent advantage is that it gives the computation a more predictable algorithmic character. Hence not only can a computation's temporal requirements be estimated, but also requirements themselves may be reduced significantly.

Effective application of this technique needs to answer two questions: what are the prerequisites for application of such techniques, and what are the possible ways of application. This paper studies both issues from an implementational point of view.

 

PM-colonies

C. Martín-Vide, Gh. PĂun

Abstract. A colony is meant to be a symbol manipulating system consisting of as simple as possible components which behave in a cooperative way such that the collective competence is strictly larger (even significantly larger) than the component competences. We introduce here colonies whose agents can only perform point mutation (hence the abbreviation PM) transformations of the common string (which represents the environment of the colony), in a vicinity of the agent. In this way, important notions of this area, such as  localization of agents, parallelism, lack of internal representation, agent interaction, come into stage in a very natural way. In contrast with the simplicity of the involved agents, the behavior of PM-colonies is quire intricate: many problems concerning the "life" of a colony are not algorithmically solvable, the number of agents in the colony or simultaneously present in the environment defines infinite hierarchies of languages, etc. Such results show that the behaviour of PM-colonies is not predictable and that their behaviour is significantly synergetic.

 

A unified approach to FAST computation of discrete sinusoidal transforms I: DCT and DST transforms

V. BRITẢÁK

Abstract. A unified approach to the fast computation of orthogonal discrete sinusoidal transforms for real data sequences is presented. Various types of discrete sinusoidal transforms (DCT) and discrete sine transform (DST are members of discrete sinusoidal transform family. The unified approach  takes advantage of a regular universal computational structure both for the DCT/DST type II (DCT-II(DST-II) and type III (DCT-III/DST-III) computation in existing real sparse matrix factorizations leading to simple, numerically stable, in  place and efficient algorithms for any N = 2m, m > 0. The computational complexity of all algorithms both in the sense of the number of arithmetic operations and structural simplicity is better or identical compared with the best known algorithms. The proposed generalized signal flow graphs are regular and confirm the importance of the universal DC-/II/DST-II (DCT/III/DST-III) computational structure for its implementation on one VLSI chip.

 

Computing the singular values of a complex matrix using one-sided Jacobi method on the intel-paragon machine

S. ROBERT

Abstract. An algorithm for computing the singular values of a complex matrix based on Rijk's improvement of the one-sided Jacobi method [12] is developed.  The aim was to find the most efficient Jacobi algorithm for the complex case retaining the same characteristics of numerical stability. Its parallelisation is given for MIMD machines and in particular the tests have been done on the Intel-Paragon machine with an estimation of a maximum of 55% gain in computation time between the classical algorithm and the improved one.


Go To  Contents