I was looking to implement this work in Nengo. It’s a deep spiking neural network that uses STDP for learning in Conv layers, Pooling layers and a classifier at the end.
The way Conv layers behave:
Each convolutional layer is comprised of several neuronal planes (similar to feature maps). Neurons in each plane are selective to the same feature and have the same synaptic weight matrix.
Neurons which fire earlier are allowed to do STDP. The winner prevents all other neurons from doing STDP. Then, it copies its updated weight matrix into other neurons in its own plane (to keep them selective to the same feature).
So my question is can I implement such behaviour and this network using NengoDL ? if so, any ideas on how to create those custom synapses ?