CODEWITHSUNDEEP
apieventsevent-handlingneural-networkbiological-neural-network
Alex Londeree
Alex Londeree

Reputation: 181

Event Driven Neural Network Simulation

I am considering building a series of event driven neural networks, inspired by actual brain architecture rather than a simple feed-forward network(For a frame of reference, it will be closer to an HTM algorithm). I'd like to simplify the neurons in the network to the point where they resemble real neurons:

-contain a series of pointers to other neurons and any necessary connection data.

-contain a value corresponding to whether its activation has passed a threshold.

-when the threshold is exceeded, it triggers an event handled by a method within the neuron, updating connection weights and updating the activation of connected neurons.

The way I see it, the event driven nature allows the network to evaluate the activation events on multiple processors, and the fact that the network is updated locally means that it does not require supervised training. I am fairly new to event driven programming (I've toyed with wxPython) so there are two questions I need to ask:

Is there a good event driven API intended for simulations, or should I just extract the features I need from something like opengl?

Do you have any suggestions as to how to limit the number of events that can occur in a given time period without skipping over events? The simulation is a chain reaction and, if not properly handled, would consume all of the processor.

Upvotes: 0

Views: 400

Answers (1)

Uri Cohen
Uri Cohen

Reputation: 3608

If I understood your idea correctly, it resemble the "integrate and fire" models. The Voltage level in the model is just your activation level.
If you want to re-use existing software for modeling such networks, you can choose from a long list at http://home.earthlink.net/~perlewitz/sftwr.html#realistic

Some of those have a python front-end, so they will probably be usable from wxPython.

Upvotes: 1

Related Questions