MSIMS@RUTGERS.ARPA (06/13/84)
From: Michael Sims <MSIMS@RUTGERS.ARPA>
[Forwarded from the Rutgers bboard by Laws@SRI-AI.]
machine learning brown bag seminar
Title: Learning Equations Solving Methods from Worked Examples
Speaker: Bernard Silver
Dept. of Artificial Intelligence
University of Edinburgh
Date: Wednesday, June 27, 1984, 12:00-1:30
Location: Hill Center, 7th floor lounge
This talk will describe LP, a program that learns new techniques for
solving equations by examining worked examples. Unlike most of the
work in this field, where the equations have been very simple, LP uses
equations of A level standard (A levels are exams taken at 18, and are
used for university selection.)
In order to be able to successfully use a new technique, LP learns
many different types of information. At the lowest level, LP compares
consecutive lines in the worked example, finding differences between
them. This allows the program to learn new rewrite rules.
LP also tries to discover the strategic purpose of each step,
expressed in terms of satisfying preconditions of following steps.
From this viewpoint, the worked example can also be considered as a
type of plan for solving the equation. LP extracts the necessary
information, and builds a plan which is stored for future use. LP
executes the plan in a flexible way to solve new equations.