Branko Souček

A parallel SVD algorithm and its application to financial ratio analysis

Abstract In this paper we describe a parallel version of the Hestenes algorithm for computing the singular value decomposition. We also describe a corresponding implementation on a transputer network. The implementation has been used to accelerate some programs for financial ratio analysis. Empirical results regarding the efficiency of our implementation are also presented.

New Preprocessing Methods for Holographic Neural Networks

We propose two new methods for preprocessing of stimulus data in holographic neural networks. The first method achieves optimal data symmetrization, and it is based on estimating the actual distributions of elements within a stimulus vector. The second method serves for data expansion, and it relies on sine and cosine functions to produce dummy stimulus elements. We describe an implementation of our methods and report on some experiments. Our proposals improve the applicability of holographic networks. Namely, symmetrization assures accuracy in reproducing learned stimulus-response associations, while expansion increases learning capacity.

Limits to growth to computerised measurements: software and algorithms

Bottleneck no 1: Software. Low microcomputer hardware costs vs high software development costs leads towards two solutions: (1) highly automated application software production; (2) the microcomputer as a receiver of standardised solutions. This paper covers the following topics: software development tools; application generators; machine-oriented high-level languages; string preprocessors; block-diagram languages. Bottleneck no 2: Algorithms. By using the proper real-time algorithm, one could improve the quality of measurement by an order of magnitude. Concrete examples are presented: computerised measurement of the impulse-response of a very large system; measurement of the probability distribution with million-channel resolution; measurement of the correlation function of an event-train. In the above examples, standard algorithms give no result. A new set of algorithms has been developed, breaking the limits of computerised measurements.