PARTIAL DOMAIN ADAPTIVE METHOD BASED ON PEER ASSISTED LEARNING CLASSIFIER

In order to solve the problem that traditional methods ignore the classifier transition scenario and further reduce the negative transfer, a partial domain adaptive method based on peer assisted learning classifier is proposed. A soft weighted maximum mean square deviation was proposed to reduce the negative transfer between the source domain and the target domain, so that the feature distribution of the source shared domain and the target domain was consistent in the feature space. A peer assisted learning method was introduced to reduce the overfitting problem of the specific target learning classifier. Experimental results on three datasets show that the proposed method not only reduces negative migration, but also solves the problem of classifier shift.