5 Criteria an agent might use in the search for a good internal model.

5.3 Cost

In a purely numerical modelling language, the semantics is a one-one correspondence, so the distinction between the form of the model (a set of parameters) and what they refer to (a set of quantities) can be identified with each other without confusion. Given the more expressive language as illustrated immediately above, this correspondence will be more complicated. For example there may be several different expressions that refer to the same set of relations in the agent's environment, but one of these may be much more efficient to use for prediction than the other. This is important for agents with limited rationality. Thus the form of models as well as the content become important.

Modelling Learning as Modelling - 23 FEB 98

[Next] [Previous] [Up] [Top] [Contents]

Generated with CERN WebMaker