We show the ROC curves for the inference of a 300-node random simplicial complex from generalized Kuramoto dynamics at three different values of l keep . Due to the exclusion of monomial terms, the ROC curves do not extend all the way to the upper right corner. Dashed lines indicate how the ROC curves would look like if we extend them by randomly adding hyperedges to the inferred structure. We see that not only is the filtering strategy able to significantly reduce computational cost without sacrificing accuracy, it also serves as a natural way to select the threshold epsilon (e.g., the end of the ROC curve for l keep = 10,000 infers over 80% of the true positives with a near 0 false positive rate).
