The rich mobile data and edge computing enabled wireless networks motivate to deploy artificial intelligence (AI) at network edge, known as \emph{edge AI}, which integrates wireless communication and machine learning. In communication, data bits are equally important, while in machine learning some data bits are more important. Therefore we can allocate more radio resources to the more important data and allocate less radio resources to the less important data, so as to efficiently utilize the limited radio resources. To this end, how to define "more or less important" of data is the key problem. In this article, we propose two importance criteria to differentiate data's importance based on their effects on machine learning, one for centralized edge machine learning and the other for distributed edge machine learning. Then, the corresponding radio resource allocation schemes are proposed to improve performance of machine learning. Extensive experiments are conducted for verifying the effectiveness of the proposed data-importance aware radio resource allocation schemes.