Table 3:

Areas under the receiver operating characteristic curve (%) using conditional likelihood as an outlier score for the task of syndrome detection. Results are not shown for the full Gaussian model with large numbers of vertices due to intractable training. Standard deviations across the cross validation folds are shown in parentheses.

	$n_{\mathrm{vert}}$
	100	1k	5k
Independent Gaussian	81.6 (2.8)	84.1 (1.8)	83.5 (2.3)
Gaussian	84.3 (3.8)
non-Gaussian	84.6 (1.5)	86.3 (1.3)	85.7 (1.6)