Multiple source domain adaptation: A bound using p-norm covering numbers

Traditional supervised learning algorithms assume that the training data and the test data are drawn from the same probability distribution. But in many cases, this assumption is too simplified, and too harsh in light of modern applications of machine learning. So domain adaptation problems are proposed when the data distribution in test domain is different from that in training domain. This paper studies the problem of domain adaptation with multiple sources, which has also received considerable attention in many areas such as natural language processing and speech processing. We introduce a novelty weighted Rademacher complexity to restrict the complexity of a hypothesis class in multiple source domain adaptation and give new generalization bounds for multiple source domain adaptation. In addition, we use an analysis of the p-norm covering number to bound our weighted Rademacher complexity which has never been discussed in multiple source domain adaptation learning.