Stephen Ratz

Graduation Semester and Year

2013

Language

English

Document Type

Thesis

Degree Name

Master of Science in Computer Science

Department

Computer Science and Engineering

First Advisor

Manfred Huber

Abstract

Without short-term memory, people would have little hope to learn and accomplish tasks. The same can be said for artificially intelligent agents. Often referred as Miller's Law, the number of working objects that a human can hold in working memory is around seven. For an AI agent, the cost of keeping additional memory blocks is exponential. Other issues to consider are what to keep in memory and for how long. Only a few of many of an agent's previous steps may be important to hold on to. This thesis project attempts to train an intelligent agent to learn what to hold onto in memory using Reinforcement Learning and Focus of Attention to accomplish a task. Function Approximation is used to mitigate the memory requirements of a task as simple as block copying. The concepts used in this thesis can be applied to any task that requires memory management.

Disciplines

Computer Sciences | Physical Sciences and Mathematics

Comments

Degree granted by The University of Texas at Arlington

Download

Share

COinS