Bidirectional convergence: A cognitive approach to generalisation
Michael K. Weir, J.G. Pohill · 2002
We present a cognitive approach to generalisation, Bidirectional Convergence. This is the implementation of a cognitive process we call concept crystallisation, whereby a concept is formed gradually from initially many possibilities which converge to a single possibility under the weight of a series of learning instances shown over a period of time. Bidirectional Convergence (BDC) is a form of concept crystallisation that represents the alternative possible concepts through. Boundary versions of the concept during learning. BDC is an abstraction of Mitchell's symbolic concept learning technique (1982). We describe how BDC is evolved from Mitchell's technique into a form suitable for incorporation into neural networks, BDC is shown to provide a best-fit to given problems.>