Fig 4. Coupling-restricted model performance and correlation time scales.

(A) Left, example of functional group of an anecdotal neuron. Diagram shows all neurons with nonzero weight (across all lags) that have directed coupling outgoing to the anecdotal neuron; these neurons constitute the functional group for the anecdotal neuron. Directionality is indicated by arrows and edge weight is denoted by thickness of the edge and arrow head size, with increasing and decreasing thickness corresponding to increasing and decreasing weight respectively. Top right, cumulative distribution functions for the percent change in test set MSE across all frames for stimulus-restricted, run-restricted, and coupling-restricted LMs using the average response stimulus model and the rotary encoder running model. (B) Cumulative distribution function of percent change in test set MSE of LM₀ and LM₁ with respect to the unrestricted LM. The exclusion of lag 0 results in a large increase in test set MSE, indicating that lag 0 edges are informative to predicting single trial responses. The exclusion of lag 1 results in a marginal increases of test set MSE, indicating that lag 1 edges are mostly uninformative. (C) Strong weight distribution segregated by lag 0 and lag 1 (“strong” weight means functional weight in top quartile of magnitudes). Strong edges are more likely to be lag 0 edges, indicating why lag 0 edges are informative to accurate predictions. (D) Probability density estimate of Pearson correlation between lag 0 and lag 1 coupling across all neurons.