Time variant net: Version #2

Mackey-Glass time variant net #2


We are now going to add so called decay neurons to our net. For more information about
decay neurons see the MemBrain help file in section 'Neurons in MemBrain' - 'Adding
Decay Neurons'.

To add the decay neurons do the following.





The net now looks as following.



Five decay neurons have been added to the input. These reflect the input signal with 'Activation Sustain
Factors ranging from 0.99 to 0.1.


We will now connect these delay neurons to the hidden neurons and the output neuron to build the
following net.



Note:

If you do not want to do the net edits yourself you can also load the example net from file:

MackeyGlassTimeVariant2.mbn


We now randomize the net and repeat the training. After some stabilization time
we get the following results:



Switching to the validation lesson during training shows the following graph.




As we can see the results did not significantly improve anymore. It looks like the
delay information already was enough for the net to learn the rules.


As a cross check we now want to delete the delay neurons from the net and keep

just the decay neurons:


Continue with time variant net version #3




Copyright © 2003 - 2007, Thomas Jetter