System State
phase
ANNEALING
T (noise)
—
E_free
—
E_nudge
—
ΔE
—
D_out (Hamming)
—
θ_mean (target)
—
Controls
Click canvas = inject heat
Auto-cool: T × 0.995 each step
Auto-cool: T × 0.995 each step
Energy / Distance Plot [Sprint 1 + 5]
E_free
E_nudge
D_out
What you're seeing
Inference: Metropolis relaxation → spins minimize E(s,J).
Nudged phase: copy of state with β·target field added → nearby equilibrium x^β.
Learning (EP-like): θ_i += (η/β)(s^nudge − s^free) on target region.
Yellow circle = output region (target = +1).
As θ grows, E_free approaches E_nudge → D_out shrinks.
Nudged phase: copy of state with β·target field added → nearby equilibrium x^β.
Learning (EP-like): θ_i += (η/β)(s^nudge − s^free) on target region.
Yellow circle = output region (target = +1).
As θ grows, E_free approaches E_nudge → D_out shrinks.