Hi all,
I’m a SNN beginner.
I’m starting to study and use the Nengo simulator
I try to use STDP learning rule to training my network
But I only can found the BCM and Oja learning part at Nengo
But I can’t found the STDP learning rule at the website
Is there any way to use STDP learning rule?
Or any complete network structure example code can let me get started quickly?
Best Regards,
Lynn
Hello,
I am working on the same thing, i don’t think there is anything ready to be used but after i did a few search, i found that the BCM rule use the pre post synaptic info and is an Unsupervised method, so as a start for now i am going trough the BCM implementation and trying to inspire from it and add a new rule for the STDP. hopefully other experienced members can confirm the approach.
Many thanks
I have few question,
When you are trying to do Unsupervised learning, how do you judge your results?
Best Regards,
Lynn
For me i switched to Nengo recently but in my old simulator that i use, i record all the spikes at the last layer for every input (digit in MNIST for example) and assign class to each neuron based on the max number of spikes, and the other spikes in the other classes are considered as loss. Sum all the losses and i get the global loss value.
Many thanks,
May i ask which simulator did you use?
Lynn.
N2S3 based on Scala and Akka actors for message exchange, in Nengo probably things are different so i still need to get used to it 
Here is a related question that you might find useful: https://forum.nengo.ai/t/any-example-using-stdp/194/2. The link to the notebook in that answer has been moved to https://github.com/nengo/nengo/blob/master/docs/examples/learning/learn-unsupervised.ipynb.
If you would like to implement a different STDP learning rule, you will have to implement it manually yourself. You can also look at this paper which asserts that BCM is equivalent to triplet-based STDP learning rules: https://www.jneurosci.org/content/26/38/9673