Can
learning mechanisms in classical neural
networks...
Neural networks are usually trained by
an iteration of the following three
steps:
- Giving an input into the network,
- Assessing the output produced by
the network,
- Readjusting synaptic weights
according to the evaluation of
the output.
Suppose the neural network in you
brain wants to learn how to make your
body perform a somersault. The input
might be defined as the muscular
activities you perfom while on the
spring-borad and in the air. The output
can be defined as the way you hit the
surface of the water. This may be
anything from a painful full-face hit to
an elegant head-on dive. The more often
you try, the better you get. Inside your
brain, the synapses of the neurons
participating in the process are
readjusted until those neurons best
contributing to a desirable result are
given their just share in shaping the
decisions of each muscular activity.
There are various mechanisms suggested
for inferring how the synapses should be
altered as a function of the assessment
of the result of the last action.
Back-propagation and forward propagation
are just two to mention.
|