WHAT WE LEARNT FROM EANN '98
(the 1998 International Conference on 
Engineering Applications of Neural Networks)


R. Parenti, Ansaldo Ricerche srl, Genova, Italy
A. Bulsari, Nonlinear Solutions Oy, Turku, Finland


It seems,  at last, that  the neural network technology is approaching
the final     phase  of real   industrial   applications. That becomes
increasingly clear from   the past EANN  conferences. Noticeably  this
year,  there  was hardly  any  talk of   justifying the  use of neural
networks  unlike in the  previous years: the  participants did not any
more sound like they were on the defensive just  because they had used
neural networks.  Moreover  it  has  been  declared many times   that,
nowadays, the biggest problems in  applications of neural networks are
no    longer related to technical   issues,  but to administrative and
managerial aspects.

The conference collected  many papers in  various fields, ranging from
biomedical engineering applications to recognition of flying aircraft.
Process engineering grew   to  be the   largest  category this   year;
moreover  a four-hour   session on  the Japanese  state-of-the-art  in
neural  network   applications   in   engineering fields    has   been
successfully  implemented. The Japanese participants presented several
interesting papers in that  session (as well as  in other sessions) on
topics  including image processing   of human faces and nuclear  power
plant monitoring.  There were several participants from industries and
a plenary session around an industrial panel  discussion was held with
panelists from Spain, Italy,  Finland, Japan and USA. They represented
electronics  and telecommunications,  process  control and  automation
companies, chemical  and  petroleum industries. The  panelists pointed
out a lot of useful considerations, some of which are reported below.

While    some   industries  are   still  practising   self-denial, the
forward-looking industries have already started using these techniques
effectively. As was pointed out in the discussion, the word "benefits"
in  the title of   the proceedings of  the  conference this year seems
timely and appropriate. Some of the industries which are still denying
the  utility of  these techniques  have had bad  experiences  in past,
typically  because of amateur attempts.  Other industries have started
deriving  benefits from  neural networks  in real  terms  in the  last
couple of years or, especially in Japan, are  strongly asking back, in
terms of real benefits, the money they had  invested in the technology
development.  People not familiar with  soft computing look upon fuzzy
logic  and neural networks as very  similar techniques, which they are
not. Fuzzy control seems to  require a smaller  change in the ways  of
thinking and  gains  easier acceptance   in the  industry. Today  many
control systems use fuzzy logic, even if in a naïve manner.

It has been pointed out that that  even PID controllers (in the 1920s)
faced  difficulties similar to  those  faced by neural networks today,
until Ziegler and Nichols came up with a detailed  and a practical way
of tuning them. The same kind of  practical approach, was said, is the
real key point needed in order to gain wide acceptance in the industry
for neural networks today. The  industries seem to want neural network
models which they can tune  themselves relatively easily.  That is not
an easy task, but a solution to it might further advance the uptake of
neural   networks   for control systems in    industries.  Some of the
European companies are working hard  in this direction, while some  of
the Japanese companies seem to focus more  on having many applications
fully developed  and  working. On   the  other  hand,  some large  USA
companies  tend to wait for reliable   technology suppliers and do not
seem to be planning hard investments on the technology itself.

One panelist pointed out that there had been no major breakthroughs in
neural  networks' basic technology in the  last ten years. It might be
partly because  all  answers are   already  available  for the   basic
questions  about neural  networks.  Very  often people  who use neural
networks to solve several  practical problems in their business,  look
upon neural   networks as just something   like a standard statistical
technique.   Most  industries need  to  increase their knowledge about
these  techniques   and gain a   better   understanding of how  neural
networks   could be used to   their  advantage. A better communication
between  the academia and industries is   desirable, but today they do
not seem to  understand each other  well. On the other hand, academics
need   to increase their knowledge   about industrial requirements and
environment, both  to be able to better  educate young engineers about
how neural  networks could be used in  the industries and to drive the
basic   research a bit  more further  towards applications.  At least,
future engineers flowing  from universities to the  industries, should
be able to easily understand when to use neural network techniques and
when to use other techniques.

For information on the EANN conferences, see 
http://www.abo.fi/~abulsari/EANN.html