Wasserstein distance-based similar adversarial network characterization model

Abstract

The invention discloses a Wasserstein distance-based similar adversarial network characterization model. Marginal probability distribution of source domain subjects and target domain subjects is reduced to the greatest extent by a method of reducing the Wasserstein distance, conditional probability distribution is reduced by a correlation enhancement method, namely, the internal relation of categories is enhanced, the scheme includes the following steps: performing sampling, filtering noise, performing mapping, setting a Wasserstein distance of a domain obfuscator, setting a gradient penalty of the domain obfuscator, adopting an association enhanced classifier, solving the similarity of feature representation from a source domain to a target domain, solving the similarity of feature representation from the target domain to the source domain, obtaining the round-trip probability of features in the source domain and the target domain, calculating the label probability of the source domain, calculating the loss of Lzw and Psts through cross entropy loss, setting the access probability, setting the target domain label probability, calculating the loss of Lop and Pv through cross entropy loss, setting the classifier loss, setting the source domain prediction classification loss, setting the number N of iterations, and when the number of times of training reaches the set number of iterations, stopping operation.

Description

technical field [0001] The invention relates to the technical field of EEG emotion classification and deep transfer learning, and in particular to a representation similarity adversarial network model based on Wasserstein distance. Background technique [0002] Compared with traditional transfer learning, deep transfer learning can directly process the original data, which can better extract features, thereby improving the accuracy of the results. It has been successfully applied in the field of brain-computer interface. Individual differences are very large, and the same individual has different differences in different time periods. Therefore, it is necessary to narrow the difference between the two fields by narrowing the marginal probability distribution and conditional probability distribution of the data of different subjects. [0003] The traditional domain adaptation method usually first extracts the features of the source domain and the target domain, and then perfo...

AI Technical Summary

Problems solved by technology

[0003] The traditional domain adaptation method usually extracts the features of the source domain and the target domain first, then performs domain adaptation on the features of the two domains, and finally uses a traditional classifier for classification. Human intervention is required in the process. The previous metric-based deep domain adaptation method is usually better than the traditional algorithm in terms of accuracy, but there are still many shortcomings in performance. For example, the accuracy rate of the domain adaptation method using MMD distance fluctuates greatly. The second-order statistical feature alignment method has limited effect on two domains with large differences, etc.

Images