Abstract
Often an agent that has to solve a problem must
choose which heuristic or strategy will help it the
most in achieving its objectives. Sometimes the
agent wishes to obtain additional units of information
on the possible heuristics and strategies in order
to choose between them, but it may be costly.
As a result, the agent's goal is to acquire enough
units of information in order to make a decision
while incurring minimal cost. We focus on situations
where the agent must decide in advance how
many units it would like to obtain. We present an
algorithm for choosing between two options, and
then formulate three methods for the general case
where there are k > 2 options to choose from. We
investigate the 2-option algorithm and the general
k-option methods effectiveness in two domains: the
3-SAT domain, and the CT computer game. In both
domains we present the experimental performance
of our models. Results will show that applying the
2-option algorithm is beneficial and provides the
agent a substantial gain. In addition, applying the
k-option method in the domains investigated results
in a moderate gain.
Original language | American English |
---|---|
Title of host publication | International Joint Conference on Artificial Intelligence |
Publisher | LAWRENCE ERLBAUM ASSOCIATES LTD |
State | Published - 2005 |