Migration learning method for heterogeneous users

Abstract

The invention discloses a migration learning method for heterogeneous users, a server and other participants. Original data cannot be obtained, and the risk of privacy disclosure is reduced to a certain extent. By domain delimitation and secondary dimension reduction screening, the correlation between the sample data and the classification target is higher, the invention can adapt to user heterogeneity, the classification effect is better, and the requirement for high classification accuracy can be met to a great extent. On the other hand, a Softmax and CNN cyclic double-classification algorithm guides unsupervised learning through supervised learning, and improves the classification accuracy of label deficiency data. Selection and delimitation of a source domain and a target domain are carried out on data obtained by multiple channels of a local end, so that enough data volume is guaranteed for transfer learning. On the basis, the requirement of multi-target output is met, and the classification accuracy is improved.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a transfer learning method for heterogeneous users. Background technique [0002] With the continuous development and maturity of traditional machine learning, it is relatively easy to train a good classification model from a large amount of labeled data. However, in real application scenarios, traditional machine learning methods still cannot fully meet the requirements. On the one hand, it is relatively difficult to obtain labeled data. Most of the data generated in life does not contain labels, and the cost of manual labeling is too high; and data collection often has to consider personal privacy and security issues, which also increases the difficulty of data acquisition. On the other hand, traditional machine learning needs to rebuild the model and train every time the data is updated, which consumes a lot of time and resources. [0003] Migration learning allevi...

AI Technical Summary

Problems solved by technology

However, in real application scenarios, traditional machine learning methods still cannot fully apply the requirements

On the one hand, it is relatively difficult to obtain labeled data

Most of the data generated in life does not contain labels, and the cost of manual labeling is too high; and data collection often has to consider personal privacy and security issues, which also increases the difficulty of data acquisition in this step

On the other hand, traditional machine learning needs to rebuild the model and train every time the data is updated, which consumes a lot of time and resources

[0003] Migration learning alleviates the data pressure of traditional machine learning to a certain extent, but it is not possible to carry out migration learning in any situation, and the effect of "migration" is also affected by many factors

Most of the current research uses random source domain data, resulting in low classification accuracy, and cannot adapt to user heterogeneity, that is, it cannot meet the needs of multi-objective classification

When using data obtained from multiple channels, due to the large difference in data correlation, the accuracy of classification results will be reduced. Randomly determining the source domain and target domain may cause transfer learning to fail to take advantage of its sufficient data volume, but the learning efficiency is not high. and accuracy cannot be guaranteed

Under the constraints of various factors, the application of transfer learning is not very extensive. Most studies only propose specific algorithms for transaction classification in a certain field, and there is no complete model architecture.

[0004] In summary, the existing classification models do not have a complete process from data collection, data processing to classification algorithms, and cannot meet the problem of multi-objective output, and the classification accuracy is difficult to guarantee

Images