Population vector

The population vector algorithm decodes intended reach direction by letting each recorded neuron cast a vote: a vector pointing along that cell's preferred direction, scaled by its current (baseline-subtracted, normalized) firing rate. Summing the votes over the population yields the population vector, which points close to the actual movement direction even though every contributing cell is only broadly tuned. Historically this was the first population read-out to run in real time and drive a cursor, and it made vivid the idea that direction lives in the ensemble rather than in any single cell.

As a decoder the population vector is only a special case of the optimal linear estimator, and it is unbiased only when preferred directions tile the sphere uniformly and every cell has equal modulation depth. Real ensembles violate both assumptions: preferred directions are clustered, modulation varies, and neurons are correlated, so the naive population vector is systematically biased toward over-represented directions. Optimal linear estimators and Kalman filters correct for this by effectively whitening against the preferred-direction distribution and the noise covariance, which is why they have largely replaced the population vector in performance-critical BCIs.