Is neuronal state (ie voltages) preserved between and within batches during SLAYER training?

[Michaeljurado42]

I am pretty sure that this would not be preserved, but I am asking just to make sure. What happens to the internal voltages within a batch and after a batch is used during Assistant training? Do the internal voltages just get reset to zero?

[bamsumit]

By default, they do not. But all slayer neurons have a parameter called persistent_state. It is by default False, but when persistent_state=True, the neurons preserve their previous states (voltages and currents for CUBA LIF).
NOTE: You have to ensure that the batch size is always the same between iterations, otherwise the previous state is ambiguous and gets reset.