Hippocampal Pattern Separation & Completion

The hippocampus solves a paradox: it must keep near-identical memories from blurring together, yet still recall a whole memory from a scrap of a cue. This simulator renders the two circuits that do it. The dentate gyrus takes two overlapping cortical input patterns and, through sparse expansive coding with a competitive winner-take-all rule, recodes them into almost uncorrelated sparse codes — pattern separation. CA3's recurrent collateral network then stores those codes as attractor states of a Hebbian associative memory and, given a noisy or partial cue, converges back to the full stored pattern step by step — pattern completion. Drag the input-overlap and sparsity sliders to see separation strengthen or fail, corrupt the recall cue and watch the attractor dynamics claw the memory back, and read the live overlap statistics at each stage of the circuit.