Title:
Model-Based Reflection for Agent Evolution
Model-Based Reflection for Agent Evolution
Authors
Murdock, J. William
Authors
Advisors
Advisors
Associated Organizations
Organizational Unit
Series
Collections
Supplementary to
Permanent Link
Abstract
Adaptability is a key characteristic of intelligence. My research explores techniques
for enabling software agents to adapt themselves as their functional requirements
change incrementally. In the domain of manufacturing, for example, a software
agent designed to assemble physical artifacts may be given a new goal of disassembling
artifacts. As another example, in the internet domain, a software agent designed to
browse some types of documents may be called upon to browse a document of another
type.
In particular, my research examines the use of reflection (an agent's knowledge and
reasoning about itself) to accomplish evolution (incremental adaptation of an agent's
capabilities). I have developed a language called TMKL (Task-Method-Knowledge
Language) that enables modeling of an agent's composition and functioning. A TMKL
model of an agent explicitly represents the tasks the agent addresses, the methods it
applies, and the knowledge it uses. TMKL models are hierarchical, i.e., they represents
tasks, methods and knowledge at multiple levels of abstraction. I have also developed
a reasoning shell called REM (Reflective Evolutionary Mind) which provides support
for the execution and evolution of agents represented in TMKL. REM employs a variety of strategies for evolving TMKL agents. Some of these
strategies are purely model-based: knowledge of composition and functioning encoded
in TMKL directly enables adaptation. REM also employs two traditional artificial
intelligence and machine learning techniques: generative planning and reinforcement
learning. The combination of model-based adaptation, generative planning, and reinforcement
learning constitutes a mechanism for re
ective agent evolution which is
capable of addressing a variety of problems to which none of these individual approaches
alone is suited. My research demonstrates the computational feasibility of
this mechanism using experiments involving a variety of intelligent software agents in
a variety of domains.
Sponsor
Date Issued
2000
Extent
181027 bytes
Resource Type
Text
Resource Subtype
Technical Report