The focusing algorithm, as with most concept learning algorithms, cannot handle disjunctive descriptions of a concept.
In our training example [small, vehicle] means small and a vehicle. It is not possible to represent a tiny or medium and a vehicle. The focusing algorithm relies on the description language within the concept space. Small in our example actually means micro or tiny or medium.
Another problem with the focusing algorithm is that the only background knowledge used is represented within the concept space trees. This can be reasonable for artificial learning situations but not for natural human learning situations. In human learning, having a wider background knowledge can lead us to discover the common elements between negative and positive instances which can help us to reach new hypotheses.
Self (1987) states that for the internal learner the importance of using background knowledge to monitor a student’s learning cannot be ignored. Self suggests that machine learning by analogy (Carbonell 1983) may be appropriate here as analogical reasoning can re-use previously gained knowledge in new situations. The basis of analogy is that if two situations have a thing or a feature in common they are likely to have other things in common.
Because the focusing algorithm has many limitations it has not been widely used as a model of human concept learning. Another proposed approach to concept learning (Easterlin and Langley, 1985) is related to conceptual clustering.
Authored by Serengul Smith
E-mail to:
serengul1@mdx.ac.uk
School of Computing Science Middlesex University
Revised: September 1998