| Abstract | One of the promises of ABM is the ability to have adaptive agents make decisions inchanging environments. Though great work has been done using adaptive agents in
ABM, more research into the theoretical understanding of these systems would be useful.
Adaptive agents have already been studied within machine learning (ML)—an area of
artificial intelligence specifically concerned with adaptation and building internal models.
The first part of this paper presents a framework for understanding ML as a component of
ABM, and describes how different ML techniques can be incorporated into some ABMs.
At the high level this framework consists of two cycles that involve evaluating input,
making decisions and then generating output. Within this generalized framework, the
ML algorithm is using the ABM as an environment and a reward generator, while the
ABM is using the ML algorithm to refine the internal models of the agents. There are
many details that must be answered before any ML technique can be incorporated into an
ABM. In this paper I start to explore some guidelines for how to more closely integrate
ABM and ML and will discuss complications that arise when combining ABM and ML
techniques. To illustrate some of these issues, I will describe an integration of a ML
technique within the El Farol Bar Problem. I will conclude with some discussion of this
integration and a look toward future research.
|
