SpikingNeuron.zeros does not clear hidden states #423

Issue type

• Bug Report
• Feature Request
• Help wanted
• Other

snnTorch version

0.9.4

Description

I found a reproducible bug in the public helper SpikingNeuron.zeros.

The helper appears intended to clear hidden-state tensors to zero, but in the current implementation it does not actually modify the passed state tensor.

From source inspection, the implementation is effectively:

@staticmethod
def zeros(*args):
    for state in args:
        state = torch.zeros_like(state)

This only rebinds the local variable state and does not mutate the caller-visible tensor, so the original hidden state remains unchanged.

Because of this, calling SpikingNeuron.zeros(...) is currently a no-op in the tested path.

Minimal code to reproduce the error/bug

import torch
import snntorch as snn
from snntorch._neurons.neurons import SpikingNeuron

lif = snn.Leaky(beta=0.5, init_hidden=True)

_ = lif(torch.ones(2, 4))
before = lif.mem.clone()

SpikingNeuron.zeros(lif.mem)

after = lif.mem.clone()

print("before_nonzero =", int((before != 0).sum().item()))
print("after_nonzero =", int((after != 0).sum().item()))
print("all_zero_after =", bool(torch.equal(after, torch.zeros_like(after))))

What is expected

Calling SpikingNeuron.zeros(state) should clear the provided hidden state tensor to zero.

What actually happens

The tensor is unchanged.

Example output in my reproduction:

before_nonzero = 8
after_nonzero = 8
all_zero_after = False

This looks like an implementation bug rather than an environment issue, because the helper currently reassigns a local variable instead of modifying the provided tensor.