hxtorch.spiking.functional.eventprop.EventPropNeuronFunction

class hxtorch.spiking.functional.eventprop.EventPropNeuronFunction(*args, **kwargs)

Bases: torch.autograd.function.Function

Define gradient using adjoint code (EventProp) from norse

__init__(*args, **kwargs)

Initialize self. See help(type(self)) for accurate signature.

Methods

backward(ctx, grad_spikes, grad_membrane, …)

Implements ‘EventProp’ for backward.

forward(ctx, input, leak, reset, threshold, …)

Forward function returning hardware data if given or otherwise integrating LIF neuron dynamics, therefore generating spikes at positions > 0.

Attributes

static backward(ctx, grad_spikes: torch.Tensor, grad_membrane: torch.Tensor, grad_current: torch.Tensor) → Tuple[Optional[torch.Tensor], …]

Implements ‘EventProp’ for backward.

Parameters

  • grad_spikes – Gradient with respect to output spikes.

  • grad_membrane – Gradient with respect to membrane trace. (Not considered in EventProp algorithm, therefore not used further)

  • grad_current – Gradient with respect to current. (Not considered in EventProp algorithm, therefore not used further)

Returns

Gradient given by adjoint function lambda_i of current.

static forward(ctx, input: torch.Tensor, leak: torch.Tensor, reset: torch.Tensor, threshold: torch.Tensor, tau_syn: torch.Tensor, tau_mem: torch.Tensor, hw_data: Optional[torch.Tensor] = None, dt: float = 1e-06) → Tuple[torch.Tensor]

Forward function returning hardware data if given or otherwise integrating LIF neuron dynamics, therefore generating spikes at positions > 0.

Parameters

  • input – Weighted input spikes in shape (2, batch, time, neurons). input[0] holds graded spikes, input[1] holds zero-tensor with same shape as placeholder to allow backpropagation of two (batch, time, neurons)-shaped tensors.

  • leak – The leak voltage as torch.Tensor.

  • reset – The reset voltage as torch.Tensor.

  • threshold – The threshold voltage as torch.Tensor.

  • tau_syn – The synaptic time constant as torch.Tensor.

  • tau_mem – The membrane time constant as torch.Tensor.

  • dt – Step width of integration.

  • hw_data – Optionally available observables from hardware.

TODO: Issue 4044. The current LIF implementation differs from

hxtorch.spiking.functional.cuba_lif_integration.

Returns

Returns the spike trains, membrane and current traces. All tensors are of shape (batch, time, neurons).