Connectionists: Which simulators support structural dynamics?

[Daniel Miner]

Hi Dan et al.,

Actually, I used it in a recent paper: http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1004759 <http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1004759> I’d argue that it does have a lot of utility for the study of circuit development and self-organization, but that’s probably a more appropriate discussion for the Brian board than Connectionists.

Best,
Daniel Miner

> On Apr 6, 2016, at 6:32 PM, Dan Goodman <d.goodman at imperial.ac.uk> wrote:
> 
> I can reply for the Brian simulator.
> 
> In Brian 1.x, we have support for "dynamic" synapses allowing you to add or delete synapses at runtime.
> 
> At the moment, in Brian 2 we don't have support for this although I'd like to add it in the future. It is already fairly straightforward to add synapses at runtime, but not to remove them. You'd have to make a copy of the whole set of synapses with some removed.
> 
> We don't have support for adding/removing neurons at runtime in either version.
> 
> In both versions, it's possible to pre-allocate an oversized model as you suggested. This is more straightforward in Brian 2 but can be done in either version.
> 
> For what it's worth, the reason we haven't put more effort into this feature is that nobody has asked for it, and to my knowledge, nobody used the feature that was available in Brian 1, which is why we didn't spend any time implementing it for Brian 2.
> 
> Dan
> 
> On 05/04/2016 23:01, Fred Rothganger wrote:
>> The NEST simulator documents the capability to create/destroy synapses
>> at runtime
>> (http://www.nest-simulator.org/py_sample/structural-plasticity-example).
>> Which other simulators support a similar capability?
>> 
>> For clarity, let's define "structural plasticity" as the ability to add
>> and remove synapses and neurons while the simulation is actively
>> running. Full support would include doing this on an HPC system
>> (multiple compute nodes) and automatically rebalancing the load.
>> 
>> Several forms of partial support are possible, including:
>> * Only the ability to add/remove synapses, but not neurons.
>> * Limitation to a single compute node or thread.
>> * Pre-allocating an oversized model, some elements of which are kept
>> idle until needed.
>> 
>> -- Fred
>> 
>> Fred Rothganger
>> Neural Computing Department
>> Sandia National Laboratories
>> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20160411/43a9120d/attachment.html>