Using a Library of Efficient Data Structures and Algorithms as a Neural Network Research Tool

Abstract In this contribution we emphasize the close relation between a general neural network and an abstract graph. Since graphs are a general data structure used in many different areas, many efficient algorithms for graph manipulation and inspection exist. These algorithms can often be used to implement neural network models very efficiently. Moreover abstract data types as sets and lists can support the development of algorithms considerably. Specifically we report here on LEDA, a publicly available C++ class library of data types and algorithms. As an example some details of the implementation of a new self-organizing network, the growing cell structures, are given.