jlubo
commented
3 months ago One possible such model is given by a Leaky Integrate-and-Fire neuron with adaptive threshold and adaptive refractoriness (which I'd therefore call ATRLIF neuron).
The dynamics of the model are described by the following formulation, originating from Lava-DL (see here and here):
-
Current: $i[t] = (1-\delta_i) \cdot i[t-1] + x[t]$
-
Voltage: $v[t] = (1-\delta_v) \cdot v[t-1] + i[t] + \mathrm{bias}$
-
Threshold: $\theta[t] = (1-\delta_{\theta})\,(\theta[t-1] - \theta_0) + \theta_0$
-
Refractory state: $r[t] = (1-\delta_r)\,r[t-1]$
-
Spike event: $s[t] = (x[t] - r[t]) \geq \theta[t]$
-
Refractory post-spike event: $r[t] = r[t] + 2\,\theta[t]$
-
Threshold post-spike event: $\theta[t] = \theta[t] + \theta_{\text{step}}$
With the following parameters (cf. the PR here):
-
$\delta_i$: Decay constant for current $i$.
-
$\delta_v$: Decay constant for voltage $v$.
-
$\delta_\theta$: Decay constant for threshold $\theta$.
-
$\delta_r$: Decay constant for refractory state $r$.
-
$\theta_0$: Initial/baseline value of threshold $\theta$.
-
$\theta_\text{step}$: Increase of threshold theta upon the occurrence of a spike.
-
$\mathrm{bias}$: Neuron's bias.
PhilippPlank
So far, neurons with firing adaptation have not yet been added to the Lava library (although according neurons have been implemented in Lava-DL, see the links below). But such neurons are needed for applications that require sophisticated neuron-intrinsic plasticity as well for realistic neuroscience models.