norse.torch.functional.lsnn

norse.torch.functional.lsnn#

Long-short term memory module, building on the work by [G. Bellec, D. Salaj, A. Subramoney, R. Legenstein, and W. Maass](IGITUGraz/LSNN-official).

The LSNN dynamics is similar to the lif equations, but it adds an adaptive term $b$ :

\begin{array}{r} \begin{aligned} \dot{v} & = 1 / τ_{mem} (v_{leak} - v + i) \\ \dot{i} & = - 1 / τ_{syn} i \\ \dot{b} & = - 1 / τ_{b} b \end{aligned} \end{array}

This adaptation is applied in the jump condition when the neuron spikes:

z = Θ (v - v_{th} + b)

Contrast this with the regular LIF jump condition:

z = Θ (v - v_{th})

In practice, this means that the LSNN neurons adapt to fire more or less given the same input. The adaptation is determined by the $τ_{b}$ time constant.

Functions

`ada_lif_step`(input_tensor, state, ...[, p, dt])	Euler integration step for LIF Neuron with adaptation.
`lsnn_feed_forward_step`(input_tensor, state)	Euler integration step for LIF Neuron with threshold adaptation.
`lsnn_step`(input_tensor, state, ...[, p, dt])	Euler integration step for LIF Neuron with threshold adaptation More specifically it implements one integration step of the following ODE

Classes

`LSNNFeedForwardState`(v, i, b)	Integration state kept for a lsnn module
`LSNNParameters`([tau_syn_inv, tau_mem_inv, ...])	Parameters of an LSNN neuron
`LSNNState`(z, v, i, b)	State of an LSNN neuron