According to our December 2007 Special Topics
analysis of research in artificial neural networks over
the past decade, the scientist at #4 by total citations
is Dr. Sabri Arik, with 25 papers cited a total of 675
times. He is also the top-ranking scientist in this
topic in terms of citations per paper, with an average
of 27 cites per paper.
In Essential
Science IndicatorsSM from
Thomson
Scientific, Dr. Arik’s record
includes 29 papers, the bulk of which are classified in the field of
Engineering, cited a total of 825 times for the period between January 1,
1997 and October 31, 2007. Dr. Arik is an Associate Professor in the
Department of Computer Engineering at Istanbul University in Turkey.
In this interview below, Dr. Arik talks
with ScienceWatch.com about his highly cited
research.
Please tell us a little about your research
and educational background.
I received my Ph.D. in Electrical-Electronics and Information Engineering
from the London South Bank University in 1997 in the area of Artificial
Neural Networks. Between 1997 and 2003 I was an Assistant Professor at the
Department of Electrical and Electronics Engineering at Istanbul
University. In 2003, I was appointed Head of the Department of Computer
Engineering for three years at the same university and I am still working
in this Department as an Associate Professor.
In recognition of my research achievement in the area of artificial neural
networks, I received the Distinguished Young Scientist Award from the
Turkish Academy of Sciences in 2002 and the Junior Science Award 2005 from
the Scientific and Technological Research Council of Turkey in 2005.
I am a member of the Technical Committee on Cellular Neural Networks and
Array Computing (CNN & AC) of the IEEE Circuits and Systems Society,
and also serving as an Associate Editor for Neurocomputing.
What first interested you in artificial neural
networks?
It all started by my joining Prof. Tavsanoglu’s Research Group at
London South Bank University in 1994 to do my Ph.D. studies. I was asked to
select a Ph.D. subject out of two possible topics. I made my choice for the
one entitled "Stability of Neural Networks." This whole field fascinated me
as I went along, and what most interested me has been the mathematical
nature of the work.
Your most-cited paper in our database is the 2000
IEEE Trans. Circuit Syst. article, "On the global asymptotic
stability of delayed cellular neural networks." Would you please walk
our readers through this paper—what were your goals, what did
you find, etc.?
In 1988, Chua and Yang proposed cellular neural networks (CNNs), which
constitute a class of recurrent and locally coupled arrays of identical
cells. The connectivity among the cells is determined by a set of
parameters called templates. Templates are the key parameters to determine
the stability properties of CNNs.
"Ultimately, the goal of research
in this area is to establish the conditions under which
an artificial neural network can solve some practical
engineering problems..."
In our most-cited paper, by employing a more general type of Lyapunov
functional and using the contradiction method, we have derived new
sufficient conditions for the asymptotic convergence of the equilibrium
point for CNNs. By exploiting these conditions, it is possible to design
stable templates for CNNs.
In fact, the viability of CNN applications strongly depends on the design
of stable templates. Furthermore, stable templates are also important for
the implementation of CNNs in the Very-Large-Scale Integration (VLSI)
technology. Our paper basically presented a general mathematical framework
for the stability and design of CNN templates.
Another of your highly cited papers is the 1998
IEEE Trans. Circuit Syst. article, "Equilibrium analysis of
delayed CNN's." Would you please talk briefly about this
paper?
When Chua and Yang introduced CNNs they proved that, for symmetric
templates, a CNN would be completely stable. However, this proof was
incomplete and contained some gaps. In 1997, Chua and Wu presented a more
rigorous proof of complete stability of CNNs with symmetric templates.
However, this new proof required the cell activation to be differentiable
and strictly increasing, which contradicts the piecewise linear cell
activation used in the original CNN model.
Therefore, the question of complete stability of CNNs with respect to
piecewise linear activation function was to be addressed. In our paper, we
presented the first mathematical criterion which enabled us to analyze the
complete stability of CNNs with symmetric or non-symmetric templates with
respect to piecewise linear activation function. We have also used this
criterion to derive new sufficient conditions for complete stability of
CNNs.
Where have you taken your work since the
publication of these papers?
We have extended our results to different models of neural networks such as
Hopfiled neural networks, Cohen-Grossberg neural networks, and
bidirectional associative neural networks. We have obtained new results for
global convergence dynamics of these classes of neural networks. These
results have also been highly cited in the recent years.
On the other hand, we have recently bought a CNN chip called a cellular
neural networks universal machine (CNN-UM) within a project funded by the
Scientific and Technological Research Council of Turkey. We are currently
carrying out some image and signal processing applications of CNNs on the
CNN-UM.
Where do you see this field going in five to ten
years?
Research in the area of artificial neural networks has rapidly increased in
recent years. There are now many researchers investigating theoretical and
application aspects of neural networks. Specifically, discovering new
mathematical methodologies and implementing neural networks in the VLSI
technology would have a significant impact on the future research on
artificial neural networks.
What should the "take-away lesson" about your work
be for the general public?
Ultimately, the goal of research in this area is to establish the
conditions under which an artificial neural network can solve some
practical engineering problems such as image processing, video signal
processing, nonlinear signal processing, modeling of biological systems and
higher brain functions and pattern recognition. Our results might help to
design neural networks that could solve such engineering problems.
Sabri Arik, Ph.D.
Department of Computer Engineering
Istanbul University
Istanbul, Turkey
Sabri Arik's most-cited paper with
153 cites to date:
Arik S, Tavsanoglu V, “On the global asymptotic
stability of delayed cellular neural networks”
IEEE Trans. Circuit Syst-I 47(4): 571-4, April
2000. Source:
Essential Science IndicatorsSM from
Thomson
Scientific.