A Hopfield network stores binary patterns as attractors via Hebb's rule: W_ij = Σ_μ ξ^μ_i ξ^μ_j / N. The network capacity is ~0.138N patterns. Given a noisy or partial input, asynchronous updates converge to the nearest stored memory, minimizing the Lyapunov energy E = -½ Σ W_ij s_i s_j.
Draw on the left grid to create patterns. Store them, add noise, then watch the network recall.