Extended Data Fig. 4: Simulation comparing performance of vine copula and GLM.
From: Specialized structure of neural population codes in parietal cortex outputs
a, The log likelihood of the data given GLM and NPvC models. b, Fit quality, defined as the fraction of deviance explained (FDE) across neurons for the GLM and the NPvC model. c, Single neuron conditional information \(I({r;S|B})\) computed for two different behavioral dependency functions (linear, left; quadratic, right) of neural activity as a function of their weight \((\alpha )\). Ground truth computed decoding directly the non-movement-modulated component of neural activity (black), information estimated through GLM (purple) and NpVC (orange) (for details, see Supplementary Note ‘Comparison of the performance of NPvC and GLM on simulated neural population data’). d, Similar to a but for the interaction information of neurons pairs. e, Similar to a but for the probability of (+,+,+) triplets computed over a set of simulations for each α. In all the panels (a–e), error bars are computed as the s.e.m. over 1,000 simulations. f, Triplet probabilities for α = 0, and α = 2. In all panels, we generated 1,000 simulations for each value of α.
