A computer is a machine that simulates logical processing.  There have been
many ways to program all kinds of machines over the years.  Many years ago,
adaptive processing was developed to process data where the output was
correlated to the desired response to train the processor.  (Much like
Pavlov's dogs, positive feedback.)  This has been used for a lot of
applications, most common the echo canceller in the telephone lines, where a
clean signal without echo is desired, requiring negative feedback.  An
example of positive feedback out of control is the squelch of microphone
near a speaker.

A few years ago, it was recognized that adaptive processing control loops
simulated a neuron and it was conjectured that we could make a machine that
would learn on its own instead of requiring programming.  The researchers
recognized the money trail and started calling adaptive processing and
training algorithms by buzz words like neural nets, insinuating a brain.  We
have learned a lot about such learning systems but they still need guidance.
You can program dumb computers, like your PC, to process information with
feedback to teach it.  This is very common in training vision systems where
it is easier to train a computer to recognize a part that is correct and
reject others that differ by telling it when it makes a proper decision
rather than understand the key features that the computer is recognizing and
programming the detailed information.

The advantage of neural nets is that they can develop the correlations and
learn to recognize the trigger indications without us knowing what to look
for in the beginning.  After sufficient data has been gathered and processed
and rules are developed, it is simpler and lower cost to extract the
triggers and rules and program a general purpose machine to look for those
signals.

The human mind is still necessary to interpret the data.  Consider the hive
monitor where it simply counts bees coming and going.  When the day starts,
no bees have left or come back.  As the day progresses, a few bees leave and
return a little later.  They recruit more bees that start making trips.  As
the day progresses and the nectar flow increases, more bees are acting as
field bees.  If the total of bees out of the hive is tracked throughout the
day, more bees should have left than returned.  If that is not the case, an
anomaly has occurred.  The machine only knows that something is different.
Man may step in and conclude that robbing is occurring and this needs to
trigger an event: Notify beekeeper that robbing is occurring at hive QX758!
The beekeeper then would need to investigate to determine what happened.  As
historical data is accumulated, the hive monitor would anticipate a certain
data track during the day; more bees have always exited than entered, the
difference fluctuates during the day with a peak that changes in time (the
beekeeper may look at the peak time and draw a conclusion that the buckwheat
is in flower because the peak occurred in the morning), at the end of the
day some bees don't return - within a normal range (if a large number don't
return another event has occurred).  Conclusions and training still depend
on the human.  If bees leave at a near normal rate but the returns are very
low, a pesticide kill may have occurred.  If a large number exit very
quickly in mid morning, a swarm may have just occurred.

Neural nets are being used where tremendous amounts of information are being
processed and sorted to detect correlation with an event of interest.  For
the beekeeper, the neural nets may help isolate those indicators to look for
with a simpler detector.  They are great for the researcher.  In 1986, my
team built an adaptive processor that performed 3.25 billion floating point
operations a second, at the time the worlds fastest computer, to detect
anomalies in 24 dimensional space.  Today such neural nets are dealing with
thousands of sensors to detect anomalies, events that need to be understood
and interpreted.

I am sure that I have exceeded the question as to what is a neural net for
most of you but I hope that I have also shown how it can help us understand
huge amounts of data.  When Jerry and other researchers have identified what
to look for, engineers like me will be happy to design it into a hive
monitor that sits at the hive entrance, monitors daylight, temperature,
precipitation, bee traffic, weight change, and even sounds of the hive and
then report the events over the wireless pager network via satellite to the
beekeepers managing thousands of colonies.

Dealing with the economics, I could design such a monitor today.
Unfortunately price is a function of quantity.  If I could sell two million,
I could sell them for $100 and make money.  I agree that such a device could
save significant labor but honey's price doesn't justify it yet.  It is
statistically lower cost to requeen annually and give up the wayward swarm.
As pollination becomes more important, the economics of such devices and
efficiency of utilization of capital invested may justify such measures.

 Raymond J. Lackey                           Sweet Pines Apiary
Honeybee Consultant   -   North American Fruit Explorers
Master Beekeeper    -    Eastern Apiculture Society/OSU
Past President    -    Long Island Beekeepers Association
Speaker -BOCES (schools) and LI Speaker's Association
~40 colonies(honey) >18 years experience on Long Island
Phone:(631)567-1936                        FAX:(631)262-8053
mail:      1260 Walnut Avenue, Bohemia, NY 11716-2176
web page: http://www.tianca.com/tianca2.html
email home: [log in to unmask]
email office: [log in to unmask]