Fig. 4: ODBAE exhibits a certain degree of robustness.

a Outlier detection performance of ODBAE across datasets with varying dimensionalities (Supplementary Data 2f). For each fixed-dimensional dataset, the model was trained 10 times, and the AUC score for outlier detection was recorded after each training run. b Impact of varying noise intensities on the performance of ODBAE (Supplementary Data 2g). Laplace-distributed noise L(0, s) was added to the dataset to simulate different noise levels. For each fixed value of s, the model was trained 10 times on the noise-augmented data, and the AUC score for outlier detection was recorded after each training run. c Outlier detection accuracy of ODBAE with architecture (L, C) on the Dry Bean Dataset (Supplementary Data 2h). For each fixed architecture (L, C), the model was trained 10 times, and the average AUC score across the 10 runs was recorded to evaluate performance. Error bars indicate mean ± standard error.