Research Of Domain Adaptation Method Based On Dictionary Learning

Inspired by human experience transfer,transfer learning uses different but related source domain knowledge to solve learning tasks in the current or target domain,and has received a lot of attention.The existing methods can be divided into two parts,single source and multiple source.The single source transfer learning methods only focus on how to reduce the distribution differences and ignore the sample bias between the source and the target domains.The multiple source domain adaptation mainly concentrates on knowledge transfer from each source domain to the target one,while pays no attention to the relationship between source domains.As a result,this paper proposes two kinds of domain adaptation method based on the dictionary learning.The specific algorithms are as follows,Firstly,we propose a novel unsupervised domain adaptation with sparse dictionary represtation(SRDA).We attempt to reconstruct the samples of the source and the target domain and learn a dictionary for each domain.The sparse representations and the dictionary are optimized together.A constraint was added on the sparse code of samples so that the source classifier can be shared with the target domain in the common sparse representation space.Finally,the knowledge of the source can be easily transferred to the target.Secondly,a multiple source domain adaptation method is proposed(multi-source domain adaption method based on parameter dictionary learning,DL?MSDA).It learns the intrinsic relationship between multiple source domains through the common parameter dictionary,and transfers such knowledge to the target domain to guide its learning.Comprehensive experimental results verify that SRDA and DL?MSDA can significantly outperform competitive methods for single source and multiple source domain adaptation.