An earthquake early warning system can provide ground motion predictions and issue warnings to target areas after an earthquake occurred. People and automatic facilities are able to receive warnings before the strong ground shakings and take actions to prevent from damages. In order to earn more warning time, the earthquake early warning system usually takes advantage of limited data to predict ground motions. However, based on traditional seismological methods the earthquake early warning system is not able to provide accurate ground motion predictions. In this paper, we summarize the current status of the earthquake early warning system in Central Weather Bureau and discussed further applications of machine learning techniques.