In applications of domain adaptation, there may exist multiple source domains, which can provide more or less complementary knowledge for pattern classification in the target domain. In order to improve the classification accuracy, a decision-level combination method is proposed for the multisource domain adaptation based on evidential reasoning. The classification results obtained from different source domains usually have different reliabilities/weights, which are calculated according to the domain-consistency. So the multiple classification results are discounted by the corresponding weights under belief functions framework, and then Dempster's rule is employed to combine these discounted results. In order to reduce errors, a neighborhood-based cautious decision making rule is developed to make the class decision depending on the combination result. The object is assigned to a singleton class if its neighborhoods can be (almost) correctly classified. Otherwise, it is cautiously committed to the disjunction of several possible classes. By doing this, we can well characterize the partial imprecision of classification, and reduce the error risk as well. A unified utility value is defined here to reflect the benefit of such classification. This cautious decision-making rule can achieve the maximum unified utility value, because partial imprecision is considered better than an error. Several real data sets are used to test the performance of proposed method, and the experimental results show that our new method can efficiently improve the classification accuracy with respect to other related combination methods.

No abstract

In the applications of domain adaptation (DA), there may exist multiple source domains, and each source domain usually provides some auxiliary information for object classification. The combination of such complementary knowledge from different source domains is helpful for improving the accuracy. We propose an evidential combination of augmented multi-source of information (ECAMI) method. The information sources are augmented at first by merging several randomly selected source domains to generate extra auxiliary information. We can obtain one piece of classification result with the assistance of each information source based on DA. Then these multiple classification results are combined by belief functions theory, which is expert at dealing with the uncertain information. Nevertheless, the classification results derived from different information sources may have different weights. The optimal weights are calculated by minimizing an given error criteria defined by the distance between the combination result and the ground truth using some training data. For each object, the augmented information sources will produce multiple classification results that will be discounted by the learnt weights under the belief functions framework. Then the combination of these discounted results is employed to make the final class decision. The effectiveness of ECAMI is evaluated with respect to some related methods based on several real data sets, and the experimental results show that ECAMI can significantly improve the classification accuracy.

No abstract

No abstract