Optimal Wireless Resource Allocation With Random Edge Graph Neural Networks

Abstract

We consider the problem of optimally allocating resources across a set of transmitters and receivers in a wireless network. The resulting optimization problem takes the form of constrained statistical learning, in which solutions can be found in a model-free manner by parameterizing the resource allocation policy. Convolutional neural networks architectures are an attractive option for parameterization, as their dimensionality is small and does not scale with network size. We introduce the random edge graph neural network (REGNN), which performs convolutions over random graphs formed by the fading interference patterns in the wireless network. The REGNN-based allocation policies are shown to retain an important permutation equivariance property that makes them amenable to transference to different networks. We further present an unsupervised model-free primal-dual learning algorithm to train the weights of the REGNN. Through numerical simulations, we demonstrate the strong performance REGNNs obtain relative to heuristic benchmarks and their transference capabilities.

Optimal Power Flow Using Graph Neural Networks

“…The desirable properties of locality and scalability can be achieved by a careful choice of model Φ(X; H). In particular, we focus on graph neural networks (GNNs) [13][14][15] to exploit their stability properties [21] that guarantee scalability [22,23].…”

A Study on Deep Learning for Latency Constraint Applications in Beyond 5G Wireless Systems

“…In [25], the authors propose an unsupervised learning strategy based ensemble model for sum-rate maximization in a fading multi-user interference channel, which outperforms the state-of-the-art methods. In [26], a random edge graph neural network is proposed to parameterize the resource allocation policy which is trained using an unsupervised model-free, primaldual learning method. In general, unsupervised techniques converge to a local optimum and thus suffer from performance degradation with time similar to supervised models.…”

Learning to Continuously Optimize Wireless Resource in Episodically Dynamic Environment

“…Deep learning has been successful in many applications such as computer vision [1], natural language processing [2], among others [3]. Recent works have also demonstrated that deep learning can be applied in communication systems, either by replacing an individual component in the system (such as signal detection [4,5], channel estimation [6,7], power allocation [8][9][10][11][12] and beamforming [13]), or by jointly optimizing the entire system [14,15], for achieving state-of-the-art performance. Specifically, deep learning is a datadriven method in which a large amount of training data is used to train a deep neural network (DNN) for a specific task (such as power control).…”