Neurons

The nodes here are deliberately minimal caricatures of real neurons: enough biology to make behaviour self-organise, not one channel more. This page steps back from any single node and lays out the neuron features the lab models, the biology behind each, and where it lives in the code. Think of it as the map from neuroscience to variant.

Each feature below reads: the biology → what BrainlessLab does with it.

Membrane integration & leak

A neuron’s membrane sums synaptic current and forgets it over time — a leaky integrator. → Every Falandays node does aᵢ ← aᵢ(1−λ) + input + recurrent each tick (leak $\lambda = 0.25$). The same leak turns injected noise into a temporally-correlated (Ornstein–Uhlenbeck) subthreshold signal — see spontaneous activity below.

Threshold & reset

Above a voltage threshold a neuron fires and its membrane resets. → A node fires when $a_i \ge \mu T_i$ and resets by subtraction ($a_i \leftarrow a_i - \mu T_i$), which preserves the overshoot rather than discarding it. See Falandays.

Homeostasis & intrinsic plasticity

Real neurons regulate their own excitability toward a firing set-point (intrinsic plasticity / synaptic scaling). → Each node carries a target $T_i$ that adapts every tick ($T_i \leftarrow \max(T_{\min}, T_i + \eta_T e_i)$), and recurrent weights follow a local homeostatic error rule. This “homeostasis as prediction” framing is the core of the Falandays model; :falandays_ablated pins the target to show what it buys you.

Dale’s law (excitatory / inhibitory)

A biological neuron is either excitatory or inhibitory — its outgoing synapses share one sign. → The sign=:dale axis fixes each node’s output sign; :falandays_extended bundles it with small-world wiring and sensory noise.

Spiking dendrites & dendritic computation

Dendrites are not passive wires. They host voltage-gated channels and generate local dendritic spikes, so a single neuron behaves like a small network of semi-independent nonlinear subunits — and those local events can gate plasticity at just the synapses on the active branch, independent of whether the soma fired.^{1, 2, 3} Dendrite-local prediction and burst signals are increasingly seen as a substrate for biologically plausible credit assignment.^{4, 5}

→ :falandays_dendritic partitions each node’s recurrent synapses across n_dendrites compartments. Each dendrite integrates its own routed current and fires its own spike; a dendritic spike sets an eligibility tag on the synapses feeding it, and the homeostatic weight update fires on presynaptic-spike or dendrite-tag — so a local dendritic event can license learning with no somatic/presynaptic spike. With eligibility_only=false the dendritic spikes also inject somatic current. This is a discrete, homeostatic caricature of dendritic computation, distinct from the biophysical :compartmental_* multi-compartment ODE cells.

simulate(:wall; node=:falandays_dendritic, ticks=300)

Source: src/nodes/Dendritic.jl (port of v0.2/crho/node_dendritic.py).

Conduction delays & spatial embedding

Signals take time to travel down axons, and neurons sit in physical space where nearby cells connect more readily. → :falandays_delayed gives per-edge conduction delays; :falandays_spatial embeds nodes in a metric space with a distance-dependent connection kernel. :falandays_hemispheric uses that embedding for two contralaterally-wired half-brains.

Connection kernel. kernel=:exp (default, $p(d) = p_0 e^{-d/\lambda}$) or kernel=:power_law ($p(d) = p_0/(1+d/d_0)^\alpha$ — bounded at $p_0$ as $d\to0$, no singularity, and a heavier tail than the exponential for $d \gg d_0$, i.e. more long-range connections at a given density). Applies to both node families.

Effector wiring. By default (effector_wiring=:bernoulli) node-effector links are a flat Bernoulli(link_p), independent of geometry. effector_wiring=:spatial anchors effectors in the same metric space (mirrored left/right for :falandays_hemispheric, so laterality is preserved) and draws their connection probability from the same kernel as recurrent wiring — a real effector, not just a node, has a location.

Hemispheric integration. callosum_density sets the probability of a bidirectional homotopic cross-hemisphere recurrent link per node pair (0.0 = fully isolated hemispheres); it is a flat coin flip, independent of kernel, and stacks with the within-hemisphere kernel-based wiring. contralateral (default true) controls whether receptor/effector wiring crosses the midline (as in real cortex) or stays same-side.

simulate(:wall; node=:falandays_spatial,
    node_kwargs=(kernel=:power_law, alpha=2.0, d0=0.3, effector_wiring=:spatial))
simulate(:wall; node=:falandays_hemispheric,
    node_kwargs=(kernel=:power_law, callosum_density=0.25, contralateral=true))

Source: src/nodes/Spatial.jl (PowerLawKernel, build_spatial_connectome, build_hemispheric_connectome); sweep axes: node.kernel, node.p0, node.lambda/node.d0/node.alpha, node.effector_wiring, node.callosum_density, node.contralateral.

Endogenous / spontaneous activity

Some cells generate spikes with no external input at all — a metabolically-powered, internally-driven fluctuation, not thermal leakage. → The Oosawa drive (:falandays_oosawa) adds target-gated membrane noise that keeps a blind network alive by variance. The biology — Oosawa’s Paramecium recordings and the Langevin/OU model of membrane-potential fluctuation — is written up in Falandays §The Oosawa drive.

References

1. London, M. & Häusser, M. (2005). Dendritic computation. Annual Review of Neuroscience 28, 503–532.
2. Poirazi, P. & Papoutsi, A. (2020). Illuminating dendritic function with computational models. Nature Reviews Neuroscience 21, 303–321.
3. Gidon, A. et al. (2020). Dendritic action potentials and computation in human layer 2/3 cortical neurons. Science 367, 83–87.
4. Urbanczik, R. & Senn, W. (2014). Learning by the dendritic prediction of somatic spiking. Neuron 81, 521–528.
5. Payeur, A., Guerguiev, J., Zenke, F., Richards, B. A. & Naud, R. (2021). Burst-dependent synaptic plasticity can coordinate learning in hierarchical circuits. Nature Neuroscience 24, 1010–1019.
6. Oosawa, F. (2007). Spontaneous activity of living cells. BioSystems 88, 191–201.