[Elecraft] Elecraft CW Net Announcement

[kevinr]

Good Evening,

    There have been a number of sunny days this week.  They are starting 
to form a pattern :)  I saw a few trilliums as I struggled through the 
thicket of hemlock along the southern edge of my land.  I did find a few 
places where the deer had made a path, but they are rather  short.  If I 
follow I have to go on my hands and knees.  I would rather find another 
way.  There are more bedding areas now, some opening up to my east as 
the alder grow taller.  Soon the deer will be more common at this 
elevation as their fodder grows.

   A tiny sunspot showed up today.  Solar flux is still low at 72 sfu 
with a weak auroral oval.  But there are more hours of sunlight now, at 
least in the northern hemisphere, providing more time to ionize the 
Heaviside layer.  I replaced the lighting in my shack which should 
improve my mood, if nothing else.  Oregon's winter offers little 
sunshine, even artificial light can make a difference.  However, a day 
spent bushwhacking through the alder is even better.


Please join us on (or near):

14050 kHz at 2200z Sunday (3 PM PDT Sunday)
   7047 kHz at 0000z Monday (5 PM PDT Sunday)

    73,
       Kevin. KD5ONS


-





I was advised to read journal papers in biology, ecology, and 
linguistics while taking courses in grammars, artificial intelligence, 
robotics, and control theory.  This week I was reading about how flies 
employ their halteres as sensors to determine angular velocity.  The 
information gathered is sent to a central pattern generator controlling 
their flight muscles.

"Central pattern generators (CPG) are neurons or neural circuits that 
produce periodic output without requiring patterned input." Artificial 
neural networks mimic neural pathways in animals. Worms and insects have 
simple neural structures making them easy to study and model.  I created 
a neural network to simulate the motion of a frightened sea slug.  In 
the case of the tritonia sea slug, a single nerve input triggers the 
generator to alternately flex the slug's dorsal and ventral muscles to 
propel it away from danger.  If the triggering enervation ceases, the 
cycle continues for a set period of time.  If the danger persists the 
generator keeps getting triggered, adding another cycle.

The pattern generator allows the organism to use its other nerves, for 
other purposes, while the CPG does its job.  If it was on a computer, 
this method would be called distributed processing.  More complex 
organisms have more generators.  A house fly has numerous generators, 
from those for flight, to those used for cleaning itself.  Avian species 
use CPGs for their songs and mating rituals.  Mammals have generators 
for gait or brachiation, as well as for digestion and other behaviors.  
Neural pathways can be enhanced through training, we call it muscle memory.

I wrote my training set using backpropagation through time to train a 
recurrent neural network.  If I had built a robotic sea slug this would 
have been its escape mechanism.  More recent studies show the pattern 
generators can be nested allowing a great deal of control through only a 
few neurons.  I expect people to use numerous trained neural networks as 
a basis for larger, nested networks.  Leave the weightings trained into 
individual networks fixed, as you train them to work in concert, 
tweaking the interlinking neurons for more complex tasks.

If your little gray cells need exercise here are some links to follow:

https://en.wikipedia.org/wiki/Halteres

https://academic.oup.com/icb/article/56/5/865/2420623

https://www.pnas.org/content/112/5/1481

https://www.biorxiv.org/content/10.1101/2020.09.15.298679v1.full

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0135011

https://jeb.biologists.org/content/222/20/jeb192054

https://d2l.ai/chapter_recurrent-neural-networks/bptt.html

https://machinelearningmastery.com/gentle-introduction-backpropagation-time/

https://www.sciencedirect.com/topics/engineering/recurrent-neural-network

https://stanford.edu/~shervine/teaching/cs-230/cheatsheet-recurrent-neural-networks

https://en.wikipedia.org/wiki/Recurrent_neural_network


Remember: neural plasticity is good but it requires you to forget at a 
certain rate.  It is the Red Queen dilemma for engineers :(