Domain-invariant Adaptation Exploiting Statistical and Geometrical Properties
کد مقاله : 1099-CFIS (R2)
نویسندگان:
مریم آذرکشت *1، فاطمه افسری2
1دانشکده کامپیوتر دانشگاه شهید باهنر کرمان
2گروه مهندسی کامپیوتر، دانشکده فنی و مهندسی، دانشگاه شهید باهنر کرمان، کرمان، ایران
چکیده مقاله:
Abstract- Domain adaptation aims to learn robust classifiers from a source data that performs well on the target data. In this paper, we propose a novel unsupervised domain adaptation method for object recognition and handwritten digit recognition applications. Our method learns two projections matrices that project the source and target domains into different low dimensional subspaces exploiting statistical and geometrical properties of data and learns a domain-invariant classifier by structural risk minimization. Extensive experiments demonstrate that our proposed method shows significant improvements in classification accuracy compared to state-of-the-art methods.
کلیدواژه ها:
Keywords: Domain Adaptation, Transfer Learning, Statistical Property, Geometrical Property, Distribution Alignment.
وضعیت : مقاله برای ارائه شفاهی پذیرفته شده است