Deep learning is a branch of machine learning that learns the high-level abstraction of data in a layered structure. Since its invention, it has been successfully applied in many image and speech processing applications. The success of deep learning depends on how big the data size is. Recently, the number of smart sensors and the Internet of Things have increased exponentially. This, in turn, has created huge traffic congestion in mobile and wireless communication networks. The available network resources need to be carefully utilized for seamless transmission of this large amount of data. Fortunately, deep learning performs very well with the big size of data. Therefore, the gap between machine learning research and advanced communication research should be narrowed down. In this article, we target an intelligent allocation of radio resources for 5G networks using deep learning. A framework consisting of a deep tree model and a long short-term memory network is proposed to predict future traffic congestion. Based on the prediction, the uplink and downlink ratio is adapted to utilize the resources optimally. Experimental results demonstrate that the proposed framework can achieve a low packet loss ratio and high throughput.

中文翻译：

---

基于深树模型的5G无线资源分配

深度学习是机器学习的一个分支，它学习分层结构中数据的高级抽象。自发明以来，它已成功应用于许多图像和语音处理应用中。深度学习的成功取决于数据量。最近，智能传感器和物联网的数量呈指数增长。反过来，这又在移动和无线通信网络中造成了巨大的流量拥塞。需要仔细利用可用的网络资源来无缝传输大量数据。幸运的是，深度学习在海量数据方面表现出色。因此，应缩小机器学习研究与高级通信研究之间的差距。在这篇文章中，我们的目标是使用深度学习为5G网络智能分配无线电资源。提出了由深树模型和长短期记忆网络组成的框架，以预测未来的流量拥塞。基于该预测，上行链路和下行链路比率适于最优化地利用资源。实验结果表明，该框架可以实现低丢包率和高吞吐量。