ABSTRACT
In the previous chapter, the application of CALM to modelling in psychology was discussed. In this chapter we will briefly discuss some practical applications of CALM, specifically, learning to recognize handwritten numerals. The model discussed in this chapter will be extended in the next, where genetic algorithms are applied to network design to reach much higher categorization and generalization scores. The model used in these two chapters is extremely limited, with a very coarse input grid and with virtually no preprocessing. Therefore, the preliminary results presented below fall short of the standards of real-world applications. They are presented only to illustrate the feasibility of the general approach to modelling and to outline various directions that could be followed to arrive at workable solutions. A recently commenced commercial project is aimed at developing real-world applications of CALM. In these models much more effort will be spent to make use of domain-specific knowledge in preprocessing and system design. To ensure a sufficient performance these models will be implemented in parallel hardware (see also Appendices B1–B5). The project will be primarily concerned with (autonomous) categorization of natural patterns. Before proceeding to discuss the model for handwritten character recognition, a few remarks will be made about two established approaches to pattern recognition and the way they are related to connectionism. We will, first, also briefly review some problems common to the recognition of both speech and handwritten characters.
