Allostatic Update Rule
The Allostatic Update Rule is a spiking neuron model that adjusts its internal target activation level over time, supporting dynamic homeostasis. Based on Falandays et al. (2021), it captures regulatory processes that balance activation around a moving set point.
Each neuron maintains:
- An activation level \(x_n\).
- A target activation level \(T_n\).
- A spiking threshold \(T'_n = 2T_n\).
The neuron updates as follows:
-
Activation decay with input
\(x_n \leftarrow \max(0,\; x_n \cdot \text{leakRate} + \sum_j w_j \cdot \text{spike}_j)\) -
Spike condition and reset
\(\text{if } x_n > T'_n \Rightarrow \text{spike}, \quad x_n \leftarrow x_n - T'_n\) -
Synaptic weight adaptation (only from spiking sources) \(w_j \leftarrow w_j - \frac{x_n - T_n}{N}\)
-
Target adaptation and threshold update \(T_n \leftarrow \max(1,\; T_n + \eta \cdot (x_n - T_n)), \quad T'_n \leftarrow 2T_n\)
This rule causes a neuron to maintain activity around its evolving target, modifying both its incoming weights and threshold accordingly.
Parameters
- Leak Rate: Proportion of activation retained after each time step. A value of 0.75 implies a 25% decay in the absence of input.
- Learning Rate: How quickly the target activation level adapts to current activity.
For all other parameters, see common neuron properties