Nodes — overview

A node is the neuron model that fills a reservoir. Every node satisfies the same contract, so any registered node builds for any task’s dimensions:

step!(r, R)            # advance one tick, return spikes/rates
effectors(r, spikes)   # map to an E-vector of length n_effectors(r)
reset!(r)              # reset runtime state
n_receptors(r); n_effectors(r)

Pick one with node=:symbol; variants() lists what’s registered.

Three families

Falandays — homeostatic, online-plastic

A leaky-integrate-and-fire reservoir that learns during the rollout (weights and per-node targets adapt each tick). Because it self-organises online, it is fair to run untrained. See base vs extended for the equations. The family: :falandays_base, :falandays_extended, :falandays_noisy, :falandays_ablated, :falandays_hemispheric, :falandays_oosawa, :falandays_dendritic, :falandays_spatial, :falandays_delayed. For the biology each variant abstracts, see Neurons.

Compartmental / CTRNN — evolved, fixed-weight

Dendrite → soma → hillock continuous-time recurrent cells with emergent/evolved weights and no online plasticity — adaptation lives in the dynamics of a fixed-weight system. An untrained compartmental node is random; it must be evolved to be a fair test. :compartmental_dense, :compartmental_structured.

:sorn — experimental

A self-organising recurrent network (Lazar/Triesch): binary threshold units with **STDP

• intrinsic plasticity + synaptic normalisation** — a mechanistically distinct route to near-critical dynamics, useful as a criticality positive-control. Not yet validated to reproduce avalanche scaling in this implementation.

Plasticity ⇒ how you prepare it

family	online plasticity	default benchmark prep
Falandays (base + variants)	yes	untrained (fair)
:sorn	yes	untrained (fair, once validated)
Compartmental	no	trained (required) — untrained is flagged not-comparable

Composing & extending

Node axes compose via kwargs — simulate(:wall; node=:falandays_base, sign=:dale, topology=:watts_strogatz, drive=OosawaDrive(...)); the named variants are preset kwarg bundles. To add a genuinely new node, see Extending it.

Source: src/core/Interfaces.jl (the generics), src/nodes/ (the families).