LPUs as singluar neurons

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

LPUs as singluar neurons

davidgr1995
Hello,

I was wondering if I could create an implementation that would allow LPUs to
act as single neurons and thus use neurokernel to model lower level
processes in the drosophila.

If this was to be the case and I was able to model the LPU as a single
neuron, then would the possibility of axon to axon connections be a
possibility as this has been shown to be the case with respect to DAN->KC
cell connections in the mushroom body.

Any Info would be awesome!

David



--
Sent from: http://neurokernel.67426.x6.nabble.com/
_______________________________________________
neurokernel-dev mailing list
[hidden email]
https://lists.columbia.edu/mailman/listinfo/neurokernel-dev

Reply | Threaded
Open this post in threaded view
|

Re: LPUs as singluar neurons

Yiyin Zhou
Hi David,

Yes, it is possible and you don't need to make LPUs to act like neurons.

The way Neurodriver defines "components" is very flexible. It can be a point neuron model, a synapse model, or a model for a compartment of a neuron, etc.. For each model you will need to specify what variables the model "accesses" (input) and what variables it "updates" (output) as well as how outputs are computed from inputs (by defining the CUDA kernel). See any model file under https://github.com/neurokernel/neurodriver/tree/master/neurokernel/LPU/NDComponents for example.

You can view a component as a node in a graph. Two components can be connected through a directed edge if the intersection between the "updates" variables of the from component and the "accesses" variables of the to component is not empty.


Now if you need to compartmentalize a neuron, basically you can define the component for compartment, component for resistance between compartments, and a graph determining the structure of the neuron's compartments. You can similarly define axo-axonal synapses (either chemical or electrical).

You, however, need to be cautious about the numerical stability, since currently Neurodriver has not yet considered numerical methods like staggered Euler for simulating multiple strongly coupled components that requires some kind of matrix inversions.

I hope this helps.

Yiyin



On Wed, Jun 27, 2018 at 10:37 AM davidgr1995 <[hidden email]> wrote:
Hello,

I was wondering if I could create an implementation that would allow LPUs to
act as single neurons and thus use neurokernel to model lower level
processes in the drosophila.

If this was to be the case and I was able to model the LPU as a single
neuron, then would the possibility of axon to axon connections be a
possibility as this has been shown to be the case with respect to DAN->KC
cell connections in the mushroom body.

Any Info would be awesome!

David



--
Sent from: http://neurokernel.67426.x6.nabble.com/
_______________________________________________
neurokernel-dev mailing list
[hidden email]
https://lists.columbia.edu/mailman/listinfo/neurokernel-dev



--
Yiyin Zhou
Postdoctoral Research Scientist
Department of Electrical Engineering
Columbia University


_______________________________________________
neurokernel-dev mailing list
[hidden email]
https://lists.columbia.edu/mailman/listinfo/neurokernel-dev
Reply | Threaded
Open this post in threaded view
|

Re: LPUs as singluar neurons

davidgr1995
Hi Yiyin,

Thank you for the info!
I will do some investigating and see if I can get what I want to work.

David



--
Sent from: http://neurokernel.67426.x6.nabble.com/
_______________________________________________
neurokernel-dev mailing list
[hidden email]
https://lists.columbia.edu/mailman/listinfo/neurokernel-dev