MGWHD-SVM: maximum weighted heteroscedastic migration learning algorithm. (3rd October 2021)

• Record Type: Journal Article
• Authors: Zhang, Min; Mo, Lianguang
• Is Part Of: International journal of computing science and mathematics. Volume 14:Number 1(2021)
• Page Start: 89
• Page End: 106
• Publication Date: 2021-10-03
• Subjects: MLWHD -- support vector machine -- SVM -- migration learning; Mathematics -- Periodicals; Computer science -- Periodicals; Mathematics -- Data processing -- Periodicals; 510.285
• Languages: English
• ISSNs: 1752-5055

Abstract

Maximum mean discrepancy (MMD) is a global measure of the distribution differences between domains at present, as a standard for effectively measuring the distribution differences between source and destination domains, however, MMD has some shortcomings in measuring the local structure and distribution differences between fields. This paper proposes a new measure: maximum local weighted heteroscedasticity discrepancy (MLWHD), this measure not only fully considers the local structure and distribution differences among fields, but also shows good adaptability to the exception points and noise, further, MLWHD was used to determine the maximum global weighted heteroscedasticity discrepancy (MGWHD), and MGWHD was embedded into the training of support vector machine (SVM). Finally, the test shows that the MGWHD method has better robustness.