Year: 2003
Type: dissertation
Abstract: Real-world domains involve objects, and if artificial agents are to interact successfully with our world, they will need to represent and reason about these objects, their attributes, and the relations that hold among them.Thus, the question of how to represent the world to the agent becomes important.This thesis compares two representations that are intended to allow for the use of reinforcement learning techniques in a domain that is essentially relational, a fully propositionalized representation, and a "deictic" one.The full propositional representation is impractical for large domains, and the hope was that a deictic representation would ameliorate many of the scaling problems.However, this thesis argues that a deictic representation raises a new set of problems that may make such a representation infeasible, even in fairly small domains.
Author: Sarah Finney
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2276710639