Active learning for inspection tool

A system comprising:

• a processor;
• memory accessible to the processor;
• instructions stored in the memory and executable by the processor to instruct the system to:
• receive labeled images;
• acquire unlabeled images;
• perform active learning by training an inspection learner using at least a portion of the labeled images to generate a trained inspection learner that, responsive to receipt of one of the unlabeled images by the trained inspection learner, outputs uncertainty information as to one or more features in the one of the unlabeled images;
• generate, by a density learner, responsive to receipt of the one of the unlabeled images, a density metric as to the one or more features in the one of the unlabeled images, wherein generating the density metric comprises computing a mean similarity of an unlabeled image to other images;
• train a typicality learner that generates a typicality metric responsive to receipt of an image;
• compute a relevance score, a density score using the density metric, a typicality score using the typicality metric, and an uncertainty score using the uncertainty information;
• based at least in part on at least one of the relevance score, the density score, the typicality score, or the uncertainty score, call for analysis of one of the labeled images or for analysis of the one of the unlabeled images;
• make a decision to call for labeling of the one of the unlabeled images based at least in part on the uncertainty information, the typicality metric, and the density metric;
• receive a label for the one of the unlabeled images;
• and
• further train the inspection learner using the label.

One or more computer-readable storage media comprising computer-executable instructions executable to instruct a computing system to:

• receive labeled images;
• acquire unlabeled images;
• perform active learning by training an inspection learner using at least a portion of the labeled images to generate a trained inspection learner that, responsive to receipt of one of the unlabeled images by the trained inspection learner, outputs uncertainty information as to one or more features in the one of the unlabeled images;
• generate, by a density learner, responsive to receipt of the one of the unlabeled images, a density metric as to the one or more features in the one of the unlabeled images, wherein generating the density metric comprises computing a mean similarity of an unlabeled image to other images;
• train a typicality learner that generates a typicality metric responsive to receipt of an image;
• compute a relevance score, a density score using the density metric, a typicality score using the typicality metric, and an uncertainty score using the uncertainty information;
• based at least in part on at least one of the relevance score, the density score, the typicality score, or the uncertainty score, call for analysis of one of the labeled images or for analysis of the one of the unlabeled images;
• based at least in part on the uncertainty information, the typicality metric, and the density metric, make a decision to call for labeling of the one of the unlabeled images;
• receive a label for the one of the unlabeled images;
• and
• further train the inspection learner using the label.