ALMA · Electric Brain — 9 Motorized Weights That Learn

A tiny perceptron network drawn as a real relay-logic circuit: 2 input bulbs feed 3 hidden relays and 1 output bulb. Each of the 9 connections is a potentiometer whose knob is turned by a small motor. Learning is literal: the perceptron rule computes an error, the motors rotate the pots, the resistances change, the currents shift. The math is exact — the wheels, glows and clicks are honest read-outs of it.
100% OURS · CANVAS 2D Stylized panel — a faithful analog of a 2→3→1 net, not a schematic of a specific PCB
Task · —
ABwantnet
Machine read-out
Epoch0
Presenting—
Output bulb—
Target—
Errors this epoch—
Accuracy—
PhaseReady
9 weights (pot angles)
—
Bulb = a value/signal · Relay = a threshold (fires when its coil current beats the spring) · Pot + motor = one learnable weight.
Task to learn
Speed 7/s
Rate η 0.30
Space learn/pause · S step · R reset.
One step = present a case, watch the relays settle, then let the motors nudge the 9 pots by the perceptron rule. Weights start randomized on Reset.