How to determine the parameter "threshold" in nengo_spa.WTAAssocMem()?


I find the nengo_spa example about Associative Memory.

I have questions about the part “CleanupWTA” in this example.

In WTA(winner take all), I know the SP with largest similarity to an ensemble, so I think the threshold is not needed in this situation.
So, why is the parameter “threshold” needed in nengo_spa.WTAAssocMem() and why threshold=0.3 in the example?


I think the threshold >= 0 is required because the WTA might not work properly with negative inputs. Setting the threshold > 0 might help in situations with noisy inputs, but is no strictly required.

Besides that is the WTA mechanism just a additional recurrent connection on top of the ThresholdingAssocMem.

1 Like

OK. Thank you!