class sinabs.layers.ALIFRecurrent(tau_mem: float | ~torch.Tensor, tau_adapt: float | ~torch.Tensor, rec_connect: ~torch.nn.modules.module.Module, tau_syn: float | ~torch.Tensor | None = None, adapt_scale: float | ~torch.Tensor = 1.8, spike_threshold: ~torch.Tensor = tensor(1.), spike_fn: ~typing.Callable = <class 'sinabs.activation.spike_generation.SingleSpike'>, reset_fn: ~typing.Callable = MembraneSubtract(subtract_value=None), surrogate_grad_fn: ~typing.Callable = SingleExponential(grad_width=0.5, grad_scale=1.0), min_v_mem: float | None = None, shape: ~torch.Size | None = None, train_alphas: bool = False, norm_input: bool = True, record_states: bool = False)[source]#

Adaptive Leaky Integrate and Fire neuron layer with recurrent connections which inherits from ALIF.

Pytorch implementation of a Long Short Term Memory SNN (LSNN) by Bellec et al., 2018:

Neuron dynamics in discrete time:

\[ \begin{align}\begin{aligned}V(t+1) = \alpha V(t) + (1-\alpha) \sum w.s(t)\\B(t+1) = b0 + \text{adapt_scale } b(t)\\b(t+1) = \rho b(t) + (1-\rho) s(t)\\\text{if } V_{mem}(t) >= B(t) \text{, then } V_{mem} \rightarrow V_{mem} - b0, b \rightarrow 0\end{aligned}\end{align} \]

where \(\alpha = e^{-1/\tau_{mem}}\), \(\rho = e^{-1/\tau_{adapt}}\) and \(w.s(t)\) is the input current for a spike s and weight w.

By default there will not be any synaptic current dynamics used. You can specify tau_syn to apply an exponential decay kernel to the input:

\[i(t+1) = \alpha_{syn} i(t) (1-\alpha_{syn}) + input\]
  • tau_mem (float | Tensor) – Membrane potential time constant.

  • tau_adapt (float | Tensor) – Spike threshold time constant.

  • rec_connect (Module) – An nn.Module which defines the recurrent connectivity, e.g. nn.Linear

  • tau_syn (float | Tensor | None) – Synaptic decay time constants. If None, no synaptic dynamics are used, which is the default.

  • adapt_scale (float | Tensor) – The amount that the spike threshold is bumped up for every spike, after which it decays back to the initial threshold.

  • spike_threshold (Tensor) – Spikes are emitted if v_mem is above that threshold. By default set to 1.0.

  • spike_fn (Callable) – Choose a Sinabs or custom torch.autograd.Function that takes a dict of states, a spike threshold and a surrogate gradient function and returns spikes. Be aware that the class itself is passed here (because torch.autograd methods are static) rather than an object instance.

  • reset_fn (Callable) – A function that defines how the membrane potential is reset after a spike.

  • surrogate_grad_fn (Callable) – Choose how to define gradients for the spiking non-linearity during the backward pass. This is a function of membrane potential.

  • min_v_mem (float | None) – Lower bound for membrane potential v_mem, clipped at every time step.

  • shape (Size | None) – Optionally initialise the layer state with given shape. If None, will be inferred from input_size.

  • train_alphas (bool) – When True, the discrete decay factor exp(-1/tau) is used for training rather than tau itself.

  • norm_input (bool) – When True, normalise input current by tau. This helps when training time constants.

  • record_states (bool) – When True, will record all internal states such as v_mem or i_syn in a dictionary attribute recordings. Default is False.

  • Input: \((Batch, Time, Channel, Height, Width)\) or \((Batch, Time, Channel)\)

  • Output: Same as input.


The membrane potential resets according to reset_fn for every spike.


This attribute is only available if tau_syn is not None.


The deviation from the original spike threshold.


The current spike threshold that gets updated with every output spike.

forward(input_data: Tensor)[source]#

input_data (Tensor) – Data to be processed. Expected shape: (batch, time, …)


Output data with same shape as input_data.