Information Technology Reference
In-Depth Information
Cellular Automata Machine for Pattern
Recognition
Pradipta Maji
1
, Niloy Ganguly
2
, Sourav Saha
1
,AnupK.Roy
1
, and
P. Pal Chaudhuri
1
1
Department of Computer Science & Technology, Bengal Engineering College (D U),
Howrah, India 711103,
{
pradipta@,sous@,anup@,ppc@
}
cs.becs.ac.in
2
Computer centre, IISWBM, Calcutta, India 700073,
n ganguly@hotmail.com
Abstract.
This paper reports a
Cellular Automata Machine
(
CAM
)
as a general purpose pattern recognizer. The
CAM
is designed around
a general class of
CA
known as
Generalized Multiple Attractor Cellu-
lar Automata
(
GMACA
). Experimental results confirm that the sparse
network of
CAM
is more powerful than conventional dense network of
Hopfield Net for memorizing unbiased patterns.
1 Introduction
This paper reports the design of a Cellular Automata Machine (CAM)toad-
dress the problem of Pattern Recognition. The design is based on an elegant
computing model of a particular class of Cellular Automata (CA) referred to
as Generalized Multiple Attractor CA (GMACA). The extensive experimental
results confirm that CAM provides an e
"
cient and cost-effective alternative to
the dense network of Neural Net for solving pattern recognition problem.
The
Associative Memory Model
provides an excellent solution to the problem
of pattern recognition. This model divides the entire state space into some pivotal
points (a, b, c, d of
Fig.1
) that represent the patterns learnt. The states close to
a pivotal point get associated with a specific learnt pattern. Identification of an
input pattern (without or with distortion due to noise) amounts to traversing the
transient path (
Fig.1
) from the given input pattern to the closest pivotal point.
As a result, the process of recognition becomes independent of the number of
patterns learnt
.
In early 80's, the seminal work of Hopfield [1] made a breakthrough by mod-
eling a
recurrent, asynchronous, neural net
as an
associative memory
system.
But, the dense network of neural net and its complex structure has partially
restricted its application. Search for alternative model around simple sparse net-
work of
Cellular Automata
(
CA
) continued [2,3,4,5,6]. The
simple, regular, mod-
ular, cascadable local neighborhood
structure of
Cellular Automata
(
CA
) serves
as an excellent sparse network model of associative memory. Such a structure
can be e
7
ciently realized with
VLSI
technology.
