A deep learning library for spiking neural networks which is based on PyTorch, focuses on fast training and supports inference on neuromorphic hardware.
It is possible to instantiate the LIF and IAF classes without a spike function. In that case the membrane potentials v_mem will be used as the layer output.
When a value for min_v_mem is provided, v_mem is clipped at this value. However, currently the clipping happens after v_mem is copied to the output tensor, hence the output can be less than the defined lower limit min_v_mem.
The fix should be easy: Move this if-else-block before the spike generation block just above it.
It is possible to instantiate the
LIFandIAFclasses without a spike function. In that case the membrane potentialsv_memwill be used as the layer output.When a value for
min_v_memis provided,v_memis clipped at this value. However, currently the clipping happens afterv_memis copied to the output tensor, hence the output can be less than the defined lower limitmin_v_mem.The fix should be easy: Move this if-else-block before the spike generation block just above it.
For the
ALIFlayer this might also be an issue.