A COMPACT GRAPHIC AND HIEROGLYPHIC SYSTEM FOR DISPLAYING SCHEMES OF DIVERSE NEURAL NETWORK CALCULATIONS

Abstract

Background. The purpose of this article is to try to eliminate ambiguities in the verbal description and graphic representations of schemes of multi-dimensional neural network calculations while reducing the volume of text and graphic information. Materials and methods. By the analogy with the standard description of Boolean logic, it is purposed to introduce a finite number of designations for the types of artificial neurons. It is also proposed to use elements of hieroglyphic image formation when graphically displaying artificial neurons (smooth ovals are continual transformations, and angular rectangles are operations for quantizing continuous data). When artificial neurons with multilevel quantization are used, their rectangular quantizers get several outputs. Results. The proposed standardization version of the alphabet of graphic symbols leads to an ambiguity’s decrease in the graphic illustrations of schemes for diverse neural network calculations and an increase in the compactness of the created illustrative materials. Conclusions. The proposed scheme for the formation of graphic and hieroglyphic illustrations is compact (does not lead to the appearance of thousands of hieroglyphs) and allows using a small number of basic images to visually illustrate the vast majority of neural network calculation schemes known to the authors.

Keywords

• classification of artificial neurons
• graphic representation of neurons
• elements of hieroglyphic records