Properties of the semiparametric pairwise model.

2017-09-19T17:24:52Z (GMT) by Jan Humplik Gašper Tkačik
<p><b>A)</b> Plot of <i>V</i> (<i>E</i>) vs <i>E</i>, i.e. the inferred nonlinearities of the semiparametric pairwise model. Curves are normalized by network size <i>N</i> and shifted along the y-axis for readability. Error bars (1 SD) denote variation over different subnetworks. The black curve is an extrapolation of the other curves to a large population size. <b>B)</b> The population size dependence of the average absolute value of the nonlinearity’s second derivative. Error bars (1 SD) denote variation over different subnetworks. <b>C)</b> Scatter plot of the couplings from a semiparametric pairwise model vs those from a pairwise model fitted to the whole population of 160 neurons. <b>D)</b> Comparison of the covariances predicted by the semiparametric pairwise model vs. those estimated from the training data. As an approximate guide for the sampling noise, covariances estimated from test data are also compared to covariances estimated from training data. Inset shows the same plot but with 10000 randomly sampled third moments <b>E</b>[<i>s</i><sub><i>i</i></sub><i>s</i><sub><i>j</i></sub><i>s</i><sub><i>k</i></sub>] such that <i>i</i> ≠ <i>j</i> ≠ <i>k</i> instead of the covariances.</p>