Small sample semi-supervised learning method and device based on pseudo label noise filtering

Abstract

The invention relates to a small sample semi-supervised learning method and device based on pseudo-label noise filtering, and the method comprises the steps: carrying out the alternative optimization of a main model and a pseudo-label filtering model, training the main model through a semi-supervised learning method, and training the pseudo-label filtering model through a noise label learning method. And the optimization of the main model and the pseudo label filtering model is alternately executed for a plurality of rounds, after each round, part of non-label data and the current pseudo label are classified into the data set with the label until the optimization is stagnated, and the optimized model is output. Compared with the prior art, the model obtained by the method has the advantages of better accuracy in the aspect of image recognition and the like.

Description

technical field [0001] The invention relates to the fields of data mining and machine learning, in particular to a small-sample semi-supervised learning method and device based on pseudo-label noise filtering. Background technique [0002] Data is one of the basic elements supporting artificial intelligence technology. Especially in the field of deep learning, the quantity and quality of data are directly related to the final generalization performance of the model. At present, most mainstream research and applications rely on the support of big data, and there are millions of labeled samples. However, with the continuous penetration of deep learning technology in various vertical fields, the specificity and specialization of target tasks are also increasing. In practical scenarios, considering the cost of expert labeling, the collected labeled data may be extremely limited. Semi-supervised learning is aimed at the scene where only some of the samples in the whole sample a...

AI Technical Summary

Problems solved by technology

Most mainstream semi-supervised learning algorithms directly use the prediction results of the main model on the unlabeled sample as the supervision information of the sample. Considering the sampling bias effect of the initial labeled data, when the supervised data sampling has bias or When the number is too small, the empirical distribution of the prediction results will not be enough to approximate the real potential sample distribution. At this time, the prediction recognition results given by the main model will contain a large amount of pseudo-label noise. If the proportion of pseudo-label noise is too high, it will affect the main model. performance

Images