Active learning method combined with labeling quality control

Abstract

The invention relates to the technical field of active learning, in particular to an active learning method combined with labeling quality control, which comprises the following steps: S1, training a model; S2, carrying out pseudo labeling on all unlabeled samples; S3, analyzing calculation and observation distance; S4, selecting to-be-labeled and to-be-checked samples based on the observation distance, selecting the to-be-labeled and to-be-checked samples based on the observation distance, labeling the to-be-labeled and to-be-checked samples, adding the labeled samples into a labeled sample set, performing additional labeling on the to-be-checked samples once, determining the label based on the existing historical label, updating the labeled sample set, and returning to the step S1 after the step is completed until the model performance can be accepted or the label budget limit is reached; according to the active learning method combined with labeling quality control, the labeling quality can be controlled while active learning is carried out.

Description

technical field [0001] The invention relates to the technical field of active learning, in particular to an active learning method combined with labeling quality control. Background technique [0002] Active learning is the process of using the model to select the most valuable samples to be labeled. During the learning process, the labeling cost required to train the model is reduced by labeling the most valuable samples successively. [0003] Usually, the active learning mode does not consider the labeling quality of the sample, that is, the labels given by the labelers are considered to be reliable. But in practice, it is inevitable for labelers to give wrong labeling results. Therefore, in practical applications, active learning methods usually imply an annotation quality control method to ensure that the quality of the obtained annotations is acceptable. Both active learning and labeling quality control work independently without collaboration, which reduces the over...

AI Technical Summary

Problems solved by technology

Both active learning and labeling quality control work independently without collaborative work, which reduces the overall labeling efficiency

Images