next up previous
Next: One Disjunct Per Up: The OMLET Learning Previous: Adjusting Membership Functions

The Training Approach

  In order to learn all the various subcategories defined in a category definition tree, we utilize a machine learning approach which is based on an assumption about human learning known as one disjunct per lesson [LehnLehn1990]. Perhaps it is easiest to understand the mechanics of our learning approach if we explain the one-disjunct-per-lesson assumption in the terminology of cognitive science. Since many of the terms in machine learning are derived from the cognitive sciences, it will not be difficult to show the similarities between our algorithm and this characterization of human learning. We will also examine some of the computational characteristics of our learning algorithm that support our choice of this approach.





Larry &
Wed Oct 18 17:48:34 EDT 1995