An interactive Hopfield associative memory on a 6 by 6 grid of bipolar neurons. Three patterns are stored as minima of an energy landscape via Hebbian weights. Load a stored pattern, add noise to corrupt it, then run recall: the network updates neuron by neuron, the energy decreases monotonically, and the state relaxes into the nearest stored memory. A sparkline shows the energy falling, and a readout reports which memory was recovered.

Load a memory, add noise, then press Recall — the energy falls as the state rolls into the nearest stored minimum.
Stored memories (click to load):
Editable neuron gridClick cells to flip neurons.
Energy E(s)
0.00
Best match