Skip to main content
. 2019 Aug 17;8(8):1241. doi: 10.3390/jcm8081241

Figure 4.

Figure 4

Test predictive performance of the Bayesian models in skin lesion classification. (a) Test prediction error of the Bayesian frameworks at different numbers of MC iterations (i.e., forward passes). The best result is achieved via Bayesian DenseNet-169 at 10 MC sampling. The shaded area around the curves shows one standard deviation. The dotted lines show the test error for the best standard (i.e., non-Bayesian) counterparts in the same color. (b) Normalized confusion matrix of the Bayesian DenseNet-169 on test data with 10 MC sampling.