Abstract :
The paper explores a variety of mappings between the neural network paradigm and the object-oriented programming model. For each of the proposed mappings the advantages and disadvantages are described with specific reference to the development time and run-time efficiency of such mappings. Experimental results are presented for both a conventional machine architecture, a Sun Microsystems´ Series 3 workstation, and an object-based computing system, the REKURSIV. These results indicate that the run-time penalty for using a compiled object-oriented programming language is not necessarily significant in neural network applications. Finally, the paper discusses, in more general terms, problems that might be encountered during the cooperation of a neural network based subsystem into a conventional knowledge based system
