Method for semi-supervised learning of structured data

Abstract

The invention discloses a method for semi-supervised learning of structured data, and the method comprises the steps: building a semi-supervised adversarial neural network model for the structured data, carrying out the preprocessing of original structured data X, and enabling the features of the original data X to be divided into a category type feature subset xCT and a numerical type feature subset xNL; the original input of the model discriminator is {x1, x2, xg}, wherein x1 is a positive integer;, xu is respectively a labeled sample and an unlabeled sample; wherein xg is a sample generatedby the generator; feature sets contained in x1 and xu are the same; inputting the class feature subset xCT of the sample into an Engineering layer; obtaining a corresponding dense embedding vector E(xCT), combining the dense embedding vector E (xCT) with the numeric feature subset xNL to obtain a sample E(xCT) + xNL with a new feature set, obtaining a normalized sample containing the new featureset by applying a BN technology, inputting the new sample into a discriminator for training, and generating a sample xg which is directly used as the input of the discriminator; the generator is composed of three layers of full-connection networks, BN is applied to the output of each layer to prevent gradient dispersion, noise serves as the output, and a production sample xg with the characteristic E(xCT) + xNL is obtained.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a method for semi-supervised learning of structured data. Background technique [0002] Semi-supervised learning problems have attracted a lot of attention in many domains, such as: anomaly detection, email archiving, etc. The raw data in many real-world applications is structured and unlabeled, but a supervised learning task requires a large amount of manually labeled data as a training set, and high-quality manually labeled data means more manpower, domain knowledge and time overhead. However, semi-supervised is a learning paradigm proposed to solve this contradiction. It can use a large amount of easily obtained unlabeled data and a small amount of manually labeled data to enhance the final performance of the classifier. Essentially, this method uses a large amount of unlabeled data to correct the assumptions learned on the labeled data. [0003] At present, many differen...

AI Technical Summary

Problems solved by technology

The raw data in many real-world applications is structured and unlabeled, but a supervised learning task requires a large amount of manually labeled data as a training set, and high-quality manually labeled data means more manpower, domain knowledge and time overhead

Images