WELD%MIT-OZ@MIT-MC.ARPA (03/12/84)
From: "Daniel S. Weld" <WELD%MIT-OZ@MIT-MC.ARPA>
[Forward from the MIT bboard by SASW@MIT-MC.]
Wednesday, March 14 4:00pm 8th floor playroom
Acquisition of Procedural Knowledge from Examples
P. M. Andreae
I will describe NODDY - a system that acquires procedures from
examples. NODDY is a variation of concept learning in which the
concepts to be learned are procedures in the form of simple robot
programs. The procedures are acquired by generalising examples
obtained by leading a robot through a sequence of steps. Three
distinct types of generalisation are involved: structure
generalisation (eg. loops and branches), event generalisation (eg. the
branching conditions), and function induction.
I will also discuss two principles that arise out
of, and are illustrated by, NODDY. I claim that these principles have
application, not only to procedure acquisition, but also to any system
that does partial matching and/or generalisation of any kind.