Next: The Full Rule Up: Imitating Analogical Reasoning Previous: Describing Diagrams

Specifying the Problem

=6cm /home/dream4/paul/teaching/ai1problemsolving/slides/evansanalogy/slidespec.eps

Now to describe rules precisely.

Now if we try to apply rule1 for example, we have a problem. The rule states that we can remove triangle1 - but C has no mention of triangle1 - we need to generalise the rule.

=6cm /home/dream4/paul/teaching/ai1problemsolving/slides/evansanalogy/slidegen.eps

But we now have an over general rule. The problem of over generalising is an important one in the area of machine learning. The following rule has abandoned the structure of the task and created a search problem.

We can control this a little by including the structure from the source:

But we are even better off if we can capture something of the structure of the target too.


paul@dream.dai.ed.ac.uk
Tue Jan 9 10:51:07 GMT 1996