|
Hi all, I am a newbie in the field, not sure if this is the right place but: I am researching if there are papers on SNNs that: a) model evolving connectivity (i.e. new connections, opposed to changing weights of existing what STDP does) and b) model reward modulated STDP with long (reward) time delays (opposed to the <50ms as in the standard STDP literature). Any feedback is greatly appreciated Thxs Ben |
|
I thought i would answer my own question since I completed some research in the meantime: Generally there are different versions of plasticity:
I found the following dissertation provides a good overview on this still young field: http://fias.uni-frankfurt.de/~savin/thesis.pdf |
|
Hi Ben, for your first point, I think you can take the network architecture to be the complete graph over all nodes and then you don't need to add new connections, you can simply block an edge by setting the weight high enough that it never gets used so, unless there's something very different about SNN architecture I don't know about, changing weights is enough. On the other question, as with most things SNN, I know nothing. However, a quick google search turned up this textbook, which seems to be reasonably recent (2002) and looks like it discusses both the questions you ask in the final few chapters. thanks for the link, I knew the resource but rediscovered some useful materials. See also my answer to my own question for what i found out in the meantime.
(Sep 02 '11 at 05:27)
Ben
|
