Performance appraisal of Learning Automata in Networks
Rohit Kumar Das
Department of IT
Assam University, Silchar
Assam, India
myown12@gmail.com
Banani Das
Department of IT
Assam University, Silchar
Assam, India
banani.das.bd@gmail.com
Sudipta Roy
Department of IT
Assam University, Silchar
Assam, India
sudipta.it@gmail.com
Abstract—In stochastic environment, uncertainties are of the
higher order to predict the future events. Learning automata
(LA) can be used in such a system where prior information is
available. These automata are capable of performing better
and better with their updating potential of input to the system
with respect to the environment. Learning automata can be
used in different networks for the proper management of the
network performances. A survey for learning automata with
networks as an environment has been reported in this paper.
This paper presents a detailed investigation on the behavior of
learning automata, various networks in which learning system
can operates.
Keywords-stochastic environment; probability; learning
automata (LA); wireless networks
I. INTRODUCTION
Stochastic environment is a part of a probability theory
where the environments are fully non-deterministic and
therefore this lead to the case of probability. In other words,
the system possesses completely unorganized events in a
random fashion hence are unpredictable or the degree of
uncertainty are of higher order like all the natural events.
Stochastic environment can be seen in business area where
the internal environment is affected by random events in the
external environment. If the system changes randomly then
the present assumptions may result insufficient for the
control system to be successfully operate which may lead to
further gathering of knowledge of the environment upon
which it is operating. Therefore learning during executing
becomes an important part for this type of systems where
additional information is required at any part of the process.
Stochastic automaton acting in this manner to improve its
performance is referred to as a learning automaton in this
paper. It can update its action probabilities which results in
reduction in the number of states in comparison with
deterministic automata.
Learning can be stated as reinforcing the existing
knowledge and leads in change of behavior. Learning cannot
be done in just a single step rather it’s a process of collecting
procedural knowledge. A learning automaton is a part of
intelligent system which has the capability to adapt the
changes in environments with unknown characteristics by a
learning process [1].
The rest of the paper is organized as: Section II provides
a brief earlier works in LA. Idea and definition for learning
automata is being given in Section III. The core working
with reinforcement scheme is elaborated in Section IV.
Section V provides the information of how learning
automata are used in different stochastic environment.
Function of automaton in various layer of network is
provided in Section VI. Finally, Section VII concludes the
paper.
II. EARLIER WORKS IN LA
The foremost learning automata models were developed
in mathematical psychology. Bush and Mosteller [2]
documented a book which contains the earlier work in this
area. A detail work has been presented by the authors of [3]
in this field. The perception of using deterministic automata
operating in random environments as models of learning is
being given by Tsetlin [4]. Varshavski and Vorontsova [5]
described the use of stochastic automata with updating of
action probabilities which results in reduction in the number
of states in comparison with deterministic automata. In their
paper they have given very innovative ideas which are used
in many research works. In [6] the author attempted to
illustrate the updating schemes which is portray through
inverse optimization problem. In [7] McLaren suggested the
concept of a growing automaton by exploring the properties
of linear updating schemes.
Learning automata to real life problems include control
of fascination columns, bioreactors, management of
manufacturing plants, pattern recognition, path planning and
action selection for autonomous mobile robots, graph
partitioning, and active vehicle suspension, path planning for
manipulators and distributed fuzzy logic processor training.
III. LEARNING AUTOMATA
Learning Automata is a part of theory of intelligent
system where the learning system interacts with the
stochastic environment by applying its input and collects the
result back to the system all the way through the feedback
mechanism which is later updated and is stored in the
updating matrix of the automata. In other words, learning
automata is a finite state mechanism which interacts with
environment and gets the optimal action back.
2014 Fourth International Conference on Communication Systems and Network Technologies
978-1-4799-3070-8/14 $31.00 © 2014 IEEE
DOI 10.1109/CSNT.2014.227
1110
2014 Fourth International Conference on Communication Systems and Network Technologies
978-1-4799-3070-8/14 $31.00 © 2014 IEEE
DOI 10.1109/CSNT.2014.227
1110