| CPC G06F 16/55 (2019.01) [G06F 18/241 (2023.01)] | 20 Claims |
|
1. A data classification method, comprising:
obtaining unlabeled images;
executing Q prediction rounds about the unlabeled images, Q is a positive integer, each of the Q prediction rounds comprising:
randomly selecting assumed inlier images among the unlabeled images;
computing a first similarity matrix comprising first similarity scores of the unlabeled images relative to the assumed inlier images; and
generating intermediate inlier-outlier predictions about the unlabeled images in one prediction round according to the first similarity matrix;
aggregating the intermediate inlier-outlier predictions about the unlabeled images generated respectively in the Q prediction rounds, to select aggregate-predicted inlier images among the unlabeled images;
computing a second similarity matrix comprising second similarity scores of the unlabeled images relative to the aggregate-predicted inlier images; and
classifying each of the unlabeled images into an inlier data set or an outlier data set according to the second similarity matrix, so as to generate inlier-outlier predictions of the unlabeled images.
|