Resources allocation (RA) is a challenging task in many fields and applications including communications and computer networks. The conventional solutions of such problems usually come with a time and memory cost, especially for massive networks such as Internet-of-Things (IoT) networks. In this paper, two RA deep network models are proposed for enabling a clustered underlay IoT deployment, where a group of IoT nodes are uploading information to a centralized gateway in their vicinity by reusing the communication channels of conventional cellular users. The RA problem is formulated as a two-dimensional matching problem, which can be expressed as a traditional linear sum assignment problem (LSAP). The two proposed models are based on the recurrent neural network (RNN). Specifically, we investigate the performance of two long short-term memory (LSTM) based architectures. The results show that the proposed techniques could be used as replacement of the well-known Hungarian algorithm for solving LSAPs due to its ability to find the solution for the problems with different sizes, high accuracy, and very fast execution time. Additionally, the results show that the obtained accuracy outperforms the state-of-the-art deep network techniques.

在许多领域和应用程序中，包括通信和计算机网络，资源分配（RA）是一项具有挑战性的任务。此类问题的常规解决方案通常会花费时间和内存，特别是对于物联网（IoT）网络等大型网络。在本文中，提出了两个RA深度网络模型，以实现集群式底层物联网部署，其中一组物联网节点通过重用常规蜂窝用户的通信信道将信息上载到附近的集中式网关。RA问题被表述为二维匹配问题，可以表示为传统的线性和分配问题（LSAP）。提出的两个模型基于递归神经网络（RNN）。特别，我们研究了两种基于长期记忆（LSTM）的体系结构的性能。结果表明，所提出的技术可以代替已知的解决LSAP的匈牙利算法，因为它能够找到大小，精度高，执行时间非常快的问题的解决方案。此外，结果表明，所获得的准确性优于最新的深度网络技术。