Can someone please explain about the brain?

187 views
Skip to first unread message

plamoni

unread,
May 16, 2013, 2:37:03 AM5/16/13
to scrip...@googlegroups.com
Hello

I know multilayer perceptron, however when I check the code it's not looks like a simple MLP. Can someone please elaborate about the functionality of the brain? What is the meaning of "target value" (shouldn't target values exist only when using Backpropagation, rather than genetic algorithm)? what is it change-sensitive synapse? what are AND/OR nodes?

Thank you!

Julian Hershey

unread,
May 29, 2013, 12:16:58 AM5/29/13
to scrip...@googlegroups.com
*perception. And as far as I know it's supposed to be based on MLP logic, but it has other features, such as "target Value," which is the raw value that the connected boxes are telling a given box to be. If we did not want to use dampening ("kp"), we would just set "output" equal to the target, but we need another variable to remind the program what value the box -wants- to go to, but not what it actually equals, and that variable is "target."

BTW, are you using the latest builds? AND/OR nodes haven't been used since we switched to MLPBrain.

Change sensitive synapse is just that; a connection with this trait compares it's input box's old output with its current output. The result is then multiplied into the parent box like normal, but the value will be very negative or positive when the input box is changing.

At least, this is all my understanding of the workings of the brain. Andrej is more knowledgeable than I, and he mainly works on the brain in his updates.

-Julian

Andrej

unread,
May 29, 2013, 1:56:07 AM5/29/13
to scrip...@googlegroups.com
Thanks Julian, that's all correct.

just to add to what Julian explained, there is no backpropagation involved here as the weights of the connections between neurons are fixed at birth and remain constant during the life of every organism.

the two main differences from a vanilla MLP is that the brains allow for recurrent connections (that is, loops) and that all neurons have some damping that smooths the changes in their activations. I can't remember exactly why I introduced this additional piece, but I think it was because I didn't like brains that were changing too quickly, as if having seizures. The damping on the changes in their values attenuates these effects, and is accomplished partly through the use of the "target" variable.

-Andrej





--
You received this message because you are subscribed to the Google Groups "Scriptbots" group.
To unsubscribe from this group and stop receiving emails from it, send an email to scriptbots+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

Reply all
Reply to author
Forward
0 new messages