Dynamic input to nodes within a network

[Mac Shine]

Hi

I'm trying to gain some intuitions about how a given network structure would respond to dynamic inputs to one (or more) of it's nodes. Is there any way that the 'activity' within a particular node can be yoked to the output of a particular algorithm (e.g. a poisson process or the output of a iterated function)?

Thanks for your help,

Mac

[jyoshimi]

Hi there,

I think the quickest way to get what you want is to use activity generators:

http://simbrain.net/Documentation/docs/Pages/Network/activityGenerator.html

To create an activity generator go the network > insert menu and select activity generator.    We don't have possoin but there are other distributions you can mess with.    There is a logistic map which is a simple iterated function.   And of course, you can make little circuits which are also basically iterated functions and send their outputs to another neuron.

Best,

- Jeff

[Mac Shine]

Great! Thanks for your help.

I'm very interested in the idea of coding up neural networks to calculate the output of iterated functions. Is there a good place for me to read up on how to do this?

Thanks!

Mac

[jyoshimi]

No reading comes to mind.   Neural networks are in a sense already iterated functions, albeit fairly complex ones.  At every update a function is computed, and the results of that are fed back in to the system for the next update.   So pressing the step button is, in a sense, iterating a function.    But you are probably thinking more about simple systems like the logistic map.   You could do something like train a backprop network to do the same thing as a logistic map, but there is no general readings on the topic I can think of.  Cheers,

- Jeff