Hi, I’m interested in spiking neural network.
I want to train spiking network using STDP.
Is there any example using STDP in Nengo??
There are a couple places to find this kind of network. The first is here:
https://github.com/nengo/nengo/blob/master/examples/learning/learn_unsupervised.ipynb
Which provides a notebook that describes the BCM implementation in Nengo (and Oja), which does STDP. That connection is more fully described in these papers:
And trevor’s thesis:
I wanted to know if we could implement this BCM learning rule for a spiking CNN? This tutorial (https://www.nengo.ai/nengo-dl/examples/spiking-mnist.html) shows how to implement spiking CNNs but it uses a supervised learning rule via the modified backprop algorithm. Is there a way to replace the learning rule somehow to incorporate unsupervised learning into the network?
We don’t currently support any online learning rules on convolutional connections. So you would have to modify the network to use dense (all-to-all) connections instead. And then you wouldn’t be using NengoDL’s sim.fit interface in that case (that’s just for offline training), you would just set up your learning rule and run it like in the standard Nengo BCM example (https://www.nengo.ai/nengo/examples/learning/learn-unsupervised.html).
The notebook mentioned in @celiasmith his answer has been moved to: nengo/learn-unsupervised.ipynb at master · nengo/nengo · GitHub