Weak label data denoising method based on regularized label propagation

Abstract

The invention discloses a noise reduction method of weak labeling data based on regularized label propagation, which comprises the following steps: 1, obtaining a required weak labeling sample data set from a crowdsourcing data platform; 2, constructing a local neighborhood structure for that weakly labeled image sample data; 3, solving the sample nearest neighbor weighted similarity matrix of thelocal neighborhood structure; 4. Constructing a noise reduction model based on system state consistency for local neighborhood structure under weak supervision; 5, solving that noise reduction modelbased on the regularize label propagation method, and achieving the noise reduction of the weakly labeled sample data set. By constructing a local neighborhood structure of weak labeling data, Considering the whole weak labeling sample data set as a system, a denoising method of weak labeling data is proposed from the viewpoint of system consistency, which realizes the denoising of weak labeling data, optimizes the labeling quality of sample data set, and does not require manual experience or expert intervention in the denoising process, effectively improves the denoising accuracy and reducesthe efficiency.

Description

technical field [0001] The invention belongs to the field of data denoising, in particular to a method for denoising weakly labeled data based on regularized label propagation. Background technique [0002] In the context of big data, sample noise is ubiquitous in real life, industrial production, and engineering applications. These noises come from a wide range of sources, including system measurement errors of sensors and errors in data processing. The impact of noisy data removal on learning problems is manifold. Directly removing noisy data is a feasible strategy when the available data scale is large. However, some samples with beneficial information may be ignored, and the assumption of independent and identical distribution will no longer hold after the data is removed. In the case of a small data sample size, the removal of noisy data may directly lead to the inability to learn the model, resulting in a waste of data resources. [0003] In the field of machine lear...

AI Technical Summary

Problems solved by technology

[0006] The technical problem to be solved by the present invention is to provide a weakly labeled data noise reduction method based on regularized label propagation for the existing noise reduction methods that require manual experience to intervene, resulting in low noise reduction accuracy and efficiency

Images