Parallel distributed processing (regular processor arrays for iterative matrix calculations and assciative memories).

Η βιβλιοθήκη διαθέτει αντίτυπο της διατριβής σε έντυπη μορφή.Διατριβή (Διδακτορική)--Πανεπιστήμιο Μακεδονίας, Θεσσαλονίκη, 1998.Περιλαμβάνει βιβλιογραφικές αναφορές (σ.198-218).002/1998The field of iterative matrix calculations finds numerous applications in a variety of scientific areas. We identify this specific range of applications by encoding a series of problems through well defined algorithms and equations. Subsequently, we propose three cases of regular processor arrays, which efficiently implement the target equations. The arrays are represented through detailed descriptions of their operation and through computational diagrams. Moreover, we employed the Ptolemy environment in order to design and simulate the processed arrays. Their correct functioning is identified and a comparison is provided through tabular representation. At the same time, a proposed methodology for creating regular processor array prototypes is presented. As regards the field of artificial neural networks, we provide an extensive survey of the most important cases of systems, which are oriented towards the efficient implementation of algorithms of the specific area. Focusing on the case of associative memories, we propose efficient solutions for both recall and learning phases, considering their underlying iterative nature. The demanding problem of the learning phase is extended with two more implementations. In the first case, we utilize methods also employed in our previous propositions, while the second case implements the problem on a commercial neurocomputer. The comparison against the best current references proves that the proposed solution remains competitive, inspite the imposed limitations