Fig. 3. Network features improve the encoding model and first order FN structure is linked to kinematic tuning.
a–c Top: monkey TY. Bottom: monkey MG. a Adding the reachFN network features to the full kinematics model improves prediction of single-unit activity (TY: p = 2.24 × 10−16 by one-sided sign-test; MG: p = 1.22 × 10−16). Each unit’s hue corresponds to average in-weight in the FN. b Performance of the full kinematics model (which contains no network feature terms) increases with average in-weight to the unit in the FN (TY: r = 0.62, Pearson correlation; MG: r = 0.54). c A scatterplot of all edge weights versus the pairwise preferred trajectory correlation. Source data are provided as a Source Data file.