Survey of Adaptive Algorithms for Intelligent Agents
Loading...
Date
2019-09
Authors
Sakenov, Nurzhan
Tyler, B. J.
Journal Title
Journal ISSN
Volume Title
Publisher
Nazarbayev University School of Engineering and Digital Sciences
Abstract
Abstract—Optimal control and decision making is essential for
a wide range of problems, such as resource optimization, realtime scheduling, making decisions based on inaccurate or incomplete data, and reasoning under uncertainty. Examples of such
problems include allocation of resources during wildfires, theater
missile defense or control of unmanned combat aerial vehicles.
One approach to these kinds of problems is adaptive intelligent
agents. When mathematically optimal exact solutions are not
suitable (for example, dynamic programming is slow, offline and
requires perfect knowledge), adaptive agents can approximate
optimality with greater speed and ability to handle uncertainty
and limited knowledge. Furthermore, using a distributed decision
making approach can improve robustness of the overall system
in question.
In this paper, we will investigate several approaches to designing intelligent agents using neural networks. Neuro-dynamic
programming (NDP) is a method of approximate dynamic programming using neural networks. Neuro-Fuzzy dynamic programming (NFDP) is a variation of NDP with incorporated fuzzy
logic. There are other methods, such as Genetic Fuzzy Trees and
Neuro-Fuzzy Inference Systems. They all differ in their ways of
handling the state space and minimizing the relevant objectives.
In this paper, we will look at several problems that people have
tried to solve using these approaches. We will discuss the types of
these problems and features that make these problems well-suited
for adaptive agents.
Description
Keywords
optimal control, adaptive agents, neuro-dynamic programming, reinforcement learning, survey