Graph Embedding based Wireless Link Scheduling with Few Training Samples

Abstract

Link scheduling in device-to-device (D2D) networks is usually formulated as a non-convex combinatorial problem, which is generally NP-hard and difficult to get the optimal solution. Traditional methods to solve this problem are mainly based on mathematical optimization techniques, where accurate channel state information (CSI), usually obtained through channel estimation and feedback, is needed.To overcome the high computational complexity of the traditional methods and eliminate the costly channel estimation stage, machine leaning (ML) has been introduced recently to address the wireless link scheduling problems. In this paper, we propose a novel graph embedding based method for link scheduling in D2D networks. We first construct a fully-connected directed graph for the D2D network, where each D2D pair is a node while interference links among D2D pairs are the edges. Then we compute a low-dimensional feature vector for each node in the graph. The graph embedding process is based on the distances of both communication and interference links, therefore without requiring the accurate CSI. By utilizing a multi-layer classifier, a scheduling strategy can be learned in a supervised manner based on the graph embedding results for each node. We also propose an unsupervised manner to train the graph embedding based method to further reinforce the scalability and generalizability and develop a K-nearest neighbor graph representation method to reduce the computational complexity.Extensive simulation demonstrates that the proposed method is near-optimal compared with the existing state-of-art methods but is with only hundreds of training samples. It is also competitive in terms of scalability and generalizability to more complicated scenarios.

AI Empowered Resource Management for Future Wireless Networks

“…However, as the resource management problem is often non-convex, the neural network is also nonconvex, and as such the optimization landscape becomes highly complicated. Due to the difficulty of optimization, unsupervised training does not always outperform the supervised one (see the comparison of Table XIII and XIV in [37]). A very recent paper also conducts a comprehensive theoretical comparison between supervised and unsupervised model [38].…”

Unsupervised Learning for Asynchronous Resource Allocation In Ad-Hoc Wireless Networks

“…Resource allocation methods, generally speaking, are usually addressed through optimization methods. Because of their non-convex nature in wireless systems, standard centralized resource allocation policies are obtained through heuristic optimization methods [1][2][3] or data-driven machine learning methods [4][5][6][7][8][9]. The latter case is seeing growing interest due to its applicability in a wide range of application scenarios and lack of reliance on explicit or accurate system modeling.…”

“…In this paper we address the asynchronous decentralized wireless resource allocation problem with a novel unsupervised learning policy-based approach. By considering the interference patterns between transmitting devices as a graph [7][8][9], we capture the asynchrony patterns via the activation of the graph edges on a highly granular time scale. From this graph representation of interference and asynchrony, we implement a decentralized learning architecture as the Aggregation Graph Neural Networks (Agg-GNNs) [19].…”

Accelerating Generalized Benders Decomposition for Wireless Resource Allocation

“…Inspired by the success of machine learning (ML) in many fields, the wireless communications community has recently turned to ML to obtain more efficient methods for resource allocation problems, such as power allocation [13]- [17], link scheduling [18], [19], and user association [20]. All aforementioned works can be classified into three different learning paradigms.…”

“…For this learning paradigm, the input/output relation of a given resource optimization problem is regarded as a black box, which is directly learned by ML techniques especially the deep neural networks (DNNs). However, this learning paradigm is effective for resource allocation with only one kind of output variables [13], [18], [19] and does not work well for MINLP problems. The second one is the reinforcement learning paradigm.…”