Fig. 4: Region and ensemble size effects on decoding accuracy.
From: Robust cortical encoding of 3D tongue shape during feeding in macaques
a Comparison of decoding accuracy from M1o and SCo populations of equal size (n = 55 neurons) for all variables. Each shape represents the mean decoding accuracy for a single variable across 6 folds of cross-validation. Notably, for all but one variable for both subjects, M1o decoding was better than SCo. Circles are kinematic variables and squares are shape variables. b Decoding accuracy (mean of all variables) as a function of M1o neuronal ensemble size. Circles are kinematic variables and squares are shape variables. Shape variable values were weighted based on the given PC’s fraction variance explained when averaging. Shaded error region represents ±1 standard deviation. c Distribution of decoding accuracy of a subsample (n = 40 per monkey) of high-performing single M1o neurons. Input to histogram was the decoding accuracy of individual variables (pooled from all 40 neurons) whose R2 was positive. Thus the distribution represents the right-tail of the performance distribution. Vertical dashed lines indicate the mean accuracy (across all variables) for ensembles of 5 and 10 M1o neurons.
