Face-Space Coding 2D: Population-Vector Decoder Simulator

This simulator gives the classic norm-based "face-space" model of face recognition a working population code, rendered as a hand-drawn 2D projection you can drag to rotate. Eight stored identities sit at fixed points in a 3D face-space; 48 tuned units — a stand-in for face-selective neurons in the fusiform face area — each respond to how close a viewed face falls to their own preferred location, and a population-vector decoder reads their combined activity back into an estimated position, which is matched to the nearest stored identity. A live bar chart tracks every unit's response on each trial. Silencing a growing fraction of the population reproduces the recognition breakdown of acquired prosopagnosia, sensory noise models a harder viewing condition, and a distinctiveness slider controls how close together the stored identities sit — the same factor that makes some real faces easier to tell apart than others.