1: \begin{abstract}

2: 		Federated learning (FL) is a promising technique that enables a large amount of edge computing devices to collaboratively train a global learning model. 

3: 		Due to privacy concerns, the raw data on devices could not be available for centralized server.

4: 		Constrained by the spectrum limitation and computation capacity, only a subset of devices can be engaged to train and transmit the trained model to centralized server for aggregation.

5: 		Since the local data distribution varies among all devices, class imbalance problem arises along with the unfavorable client selection, resulting in a slow converge rate of the global model.

6: 		In this paper, an estimation scheme is designed to reveal the class distribution without the awareness of raw data.

7: 		Based on the scheme, a device selection algorithm towards minimal class imbalance is proposed, thus can improve the convergence performance of the global model.

8: 		Simulation results demonstrate the effectiveness of the proposed algorithm.

9: 	\end{abstract}

10: