Synaptic Scaling 2D: Hebbian Plasticity vs Homeostatic Scaling

Network Activity

Simulation speed 1.0x

Live Readout

Firing rate

—

Target rate

—

Homeostatic scale factor (cumulative)

—

Weight spread (max / min)

—

How it works

This is the 2D counterpart to the 3D synaptic-scaling-homeostatic-plasticity-lab scene. The 3D scene visualizes only the slow homeostatic side of the story; this version adds the fast side it only described — a genuine per-synapse rate-covariance Hebbian update running every frame, tugging against the same slow multiplicative homeostatic scale factor:

Δw_i/dt = η · pre_i(t) · (rate_inst − rate_target) / rate_target   (fast, per-synapse)
w_i ← w_i · exp( −g · (rate_avg − rate_target)/rate_target · α )   (slow, applied to ALL synapses)

Eight synapses each carry a fixed correlation with a shared background "stimulus" signal, from strongly anti-correlated to strongly correlated. When Hebbian learning is on, a synapse whose presynaptic drive tracks the postsynaptic rate gets potentiated when the neuron fires above target and depressed when it fires below — a rate-based covariance rule. Independently, the slow homeostatic loop measures the neuron's own running-average firing rate against its setpoint and nudges every synapse by the same multiplicative factor each step, which is why the bars all grow or shrink together while keeping their relative proportions.

Network activity — scales the presynaptic drive reaching every synapse, exactly like the 3D scene's activity presets.
Hebbian learning — toggles the fast, per-synapse covariance rule on or off; turn it off to see only homeostatic scaling acting on frozen relative weights.
Homeostatic scaling — toggles the slow multiplicative renormalization; turn it off with Hebbian left on to watch weights drift and the firing rate run away from its setpoint.

Running rate Instant rate