Line Plot from Scientific Research

We fit a generalized additive model (GAM) to the model neurons. Left: Model neurons were tuned to different task variables including latents (despite not explicitly training to learn them) (compare with Fig. 3 ). Right: Coupling was weak and concentrated around zero (top) but nonetheless affected the goodness of fit (bottom-right) (compare with Fig. 4 ). Model neurons exhibited mixed selectivity to a degree that was comparable to PPC neurons (bottom-left, “Methods”). Error bars denote ± 1 SEM ( n = 244/ n = 100 for data/model).

DOI