4.4 Comparison of average AUC by LAD with different Laplacians.
For each dataset, the numbers in the parentheses indicate the ranks
of each Laplacian. Average is the average AUC of each Laplacian
across all the datasets respectively. . . . . . . . . . . . . . . . . . . 112
4.5 Comparison of average AUC by FDD with different Laplacians.
For each dataset, the numbers in the parentheses indicate the ranks
of each Laplacian. Average is the average AUC of each Laplacian
across all the datasets respectively. . . . . . . . . . . . . . . . . . . 114
4.6 Comparison of AUC between full and fast version of LAD and FDD. 118
4.7 Comparison of running time (in seconds). . . . . . . . . . . . . . . 119
5.1 Clustering results of synthetic dataset in Figure 5.1. The size of selected
feature subset is 4 for all the five feature selection algorithms.
We run each algorithm 30 times on the dataset with all instances
(including normal and noisy instances), and also on the subset of
the normal instances (without any noisy instance). We report the
average NMI score only on the normal instances. . . . . . . . . . . 124
5.2 Statistics of experimental datasets. . . . . . . . . . . . . . . . . . . 140