Deep transfer learning method based on big data cross-domain analysis

Abstract

The present invention provides a deep migration learning method for cross-domain analysis of big data, by determining the value of the loss function of the deep network according to the joint distribution difference and the classification error rate, wherein the joint distribution difference is in all task-related layers corresponding to the source domain The distribution difference between the features of the feature and the joint probability distribution between the labels and the features in all task-related layers corresponding to the target domain and the joint probability distribution between the labels; and based on the value of the loss function, the parameters of the deep network are updated to Adapt the deep network to the target domain; thus, in the transfer learning process of the deep network, the joint distribution difference is used as a component of the value of the loss function of the deep network, and by updating the parameters of the deep network, while ensuring the accuracy of the source domain Realizing the matching of the joint distribution of the source domain and the target domain improves the migration ability of all task-related layers, thus bringing better results to the transfer learning of deep networks between different domains.

Description

technical field [0001] The present invention relates to the technical field of computer data analysis, and more specifically, to a deep transfer learning method for cross-domain analysis of big data. Background technique [0002] Supervised training of machine learning on a large amount of labeled data can achieve good performance and results. However, large labeled data sets are limited in number and application fields, and manual labeling of sufficient training data often requires high costs. Therefore, when faced with a target task where labeled data is scarce, how to use the existing labeled data in the source domain that is related to the target domain but obeys a different probability distribution to build an effective learner has a strong practical demand. [0003] The paradigm of learning discriminative models when there is a distributional shift between source and target domain data is called transfer learning. Migration learning attempts to build a learner that ca...

AI Technical Summary

Problems solved by technology

However, for the upper layer of the deep network, its migration ability will decrease significantly with the expansion of the differences between domains, and the upper layer of the deep network is often a task-related layer used to fit the characteristics of the sample. The decline in the migration ability of the task-related layer will Affects the effect of deep network transfer learning between different fields

Images