Can we define a block without neurons?

[fangwei123456]

In deep SNNs, we may use the structure like 'Conv-Neuron-Pool-Conv-Neuron'. In such a case, we use the pooling layer without spiking neurons. But I find that in lava.lib.dl.slayer.block.cuba.Pool, we also need to include the spiking neurons in pooling layer. So, my question is can we define a block without neurons?

[bamsumit]

slayer.block is a convenient method that combines synapses and neurons. Typically in neuromorphic hardware, you need to have neurons. That is why the block interface provides both neuron and synapse combined. However, if you need only the synapse, you can directly call slayer.synapse.Pool and it should work.

[fangwei123456]

Thanks for your reply! Another question is how can I convert lava.lib.dl.slayer.synapse.layer.Pool to Processes in lava? I find that Supported layer types does not contain a pool without neuron.

[bamsumit]

You are right. NetX network description does not support such structure nor the Loihi hardware. Therefore, we have no plans of supporting such a structure. If you really want to do it, you can break the network into two NetX halves and connect them with a custom lava process for pooling.

[fangwei123456]

Thanks!