Scaling Laws for Linear Neural Population Decoders in Macaque Motor Cortex: Neuron Count, Training Data, and Dimensionality

This study investigates scaling behavior of linear population decoders for motor cortex neural spike data as a function of simultaneously recorded neuron count and training set size within the context of computational neuroscience and neural data analysis, an area of growing scientific importance given its implications for brain-computer interface design principles and optimal electrode array sizing for motor prosthetics. Using linear discriminant analysis and partial least squares decoders trained on binned spike count population vectors with leave-one-out cross-validation, we examine neural population activity dimensionality increasing with recorded unit count, enabling linear decoder accuracy gains that plateau per a log-linear model in recordings from 6 macaque M1/PMd sites (64-188 simultaneously recorded units per session, 2,400 trial epochs total) drawn from chronic multi-electrode array recordings in macaque motor cortex during center-out reaching task performance. Results indicate that linear decoder accuracy for 8-direction reaching discrimination scales from 68.4% at n=10 neurons to 94.2% at n=180 neurons, following a diminishing-returns curve well approximated by a log-linear model (r2=0.97) (p < 0.001), with 94.2% decoding accuracy at n=180 neurons as the primary quantitative benchmark. Concordance between primary and confirmatory measurement approaches exceeded 93%, validating the analytical framework. These findings contribute empirically to computational neuroscience and neural data analysis and carry actionable implications for the design of programs and policies targeting brain-computer interface design principles and optimal electrode array sizing for motor prosthetics.