Tutorial 4: Synaptic neuron not matching expected results

[jeshraghian]

Discussed in #319

^{Originally posted by prntechnologies April 25, 2024}
I've been going through all of the tutorials and I'm having an issue with Tutorial #4. It seems that I'm not getting the same results using the Synaptic neural model as what is described in the tutorial?

Anyone else having issues following along, or is it just me?

[MoerAI]

I also confirmed that there was a problem. The output of Synaptic Conductance-based Neuron Model With Input Spikes is strange. If you modify the hyperparameters alpha and beta defined in Markdown 1.2, you can get the spike values. I'm working on translating the SNN Torch tutorials into Korean, and would appreciate it if anyone could let me know here if this issue is resolved.

[mpgl]

Hi,

I am also having the same issue with Tutorial 4. When I manually implement the synaptic LIF model as described in the tutorial it works as expected. But when I try to reproduce the same simulation with snn.Synaptic, the membrane voltage for some reason goes to infinity. I also noticed that the input spikes are not converted into synaptic current. Are there any updates regarding this issue?

Thank you!

snntorch version: 0.9.1
pytorch version: 2.3.1+cu121

Here are the two versions of my code and the simulation results:

def LIF_second_order(voltage, i_synaptic, X, w, alpha, beta, threshold=1):
  spike = (voltage > threshold)
  i_synaptic = alpha * i_synaptic + X * w
  voltage = beta * voltage + i_synaptic - spike*threshold
  return spike, i_synaptic, voltage

# Simulation parameters
num_steps = 200
alpha = 0.9
beta = 0.8
w = 0.2

# Create input and monitor lists
X = torch.zeros(num_steps)
X[::10] = 1
v_monitor = []
spike_monitor = []
isyn_monitor = []

# Run simulation
voltage = torch.zeros(1) # Initialize voltage
isynaptic = torch.zeros(1) # Initialize synaptic current
for step in range(num_steps):
    spike, isynaptic, voltage = LIF_second_order(voltage, isynaptic, X[step], w, alpha, beta)
    spike_monitor.append(spike)
    v_monitor.append(voltage)
    isyn_monitor.append(isynaptic)

# Plot results
fig, axes = plt.subplots(3, 1, sharex=True)
ax0, ax1, ax2 = axes

ax0.plot(torch.nonzero(X), torch.ones_like(torch.nonzero(X)), 'k|')
ax0.set_yticks([])
ax0.set_ylabel('Input Spikes')

ax1.plot(isyn_monitor)
ax1.set_ylabel('Synaptic Current')

ax2.plot(v_monitor)
ax2.set_ylabel('Membrane Voltage')
ax2.set_xlabel('Time')
plt.plot()

# Simulation parameters
num_steps = 200
alpha = 0.9
beta = 0.8
w = 0.2

# Create input and monitor lists
X = torch.zeros(num_steps)
X[::10] = w
v_monitor = []
spike_monitor = []
isyn_monitor = []

# Create synaptic LIF neuron
synaptic_lif = snn.Synaptic(alpha=alpha, beta=beta)

# Run simulation
isynaptic, voltage = synaptic_lif.init_synaptic()
for step in range(num_steps):
    spike, isynaptic, voltage = synaptic_lif(X[step], isynaptic, voltage)
    spike_monitor.append(spike)
    v_monitor.append(voltage)
    isyn_monitor.append(isynaptic)

# Plot results
fig, axes = plt.subplots(3, 1, sharex=True)
ax0, ax1, ax2 = axes

ax0.plot(torch.nonzero(X), torch.ones_like(torch.nonzero(X)), 'k|')
ax0.set_yticks([])
ax0.set_ylabel('Input Spikes')

ax1.plot(isyn_monitor)
ax1.set_ylabel('Synaptic Current')

ax2.plot(v_monitor)
ax2.set_ylabel('Membrane Voltage')
ax2.set_xlabel('Time')
plt.plot()