Graph Neural Networks for Wireless Communications: From Theory to Practice

Shen, Yifei; Zhang, Jun; Song, Shenghui; Letaief, Khaled B.

doi:10.1109/twc.2022.3219840

Cited by 59 publications

(23 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The generic NNs are not specific to a particular task or dataset, but can be applied for general tasks and datasets to achieve an acceptable performance. For example, with tremendous training data and large NN architecture, MLP is usually adopted as a benchmark algorithm for performance comparison in multi-input multi-output (MIMO) detection [32], power control [33], semantic communication [34], etc. However, the flexibility comes with the cost of poor data efficiency (high training overhead), poor robustness and poor generalization ability.…”

Section: B Machine Learning In 6gmentioning

confidence: 99%

“…Inspired by the pertinent COAs, some of the MOAs in the "supervised learning" category enable theoretical analysis by constructing a relationship of performance between these two types of methods. If equivalence can be proved, the performance analysis of MOAs can be developed based on that of COAs (e.g., the performance analysis of algorithm unrolling approaches [14], LBB approach [15], GNN approach [33], etc. ).…”

Section: B Machine Learning In 6gmentioning

confidence: 99%

“…For example, compared with MLP, the parameters of GNN can be shared for graphs with varying sizes and permuted input, thereby achieving good generalizability, while an MLP has to be retrained when the topology is changed. More precisely, authors in [33] theoretically showed that the GNN's generalization error and required number of training samples are O(n) and O(n 2 ) lower than those of MLPs, where n is the number of nodes in the graph.…”

Section: A Principles Of Graph Neural Networkmentioning

confidence: 99%

“…To improve the learning ability of GNN over different graphs, the aggregation function f 1 , the combination function f 2 and the pooling function PL should be carefully designed. Authors in [33] gave general guidelines for designing these functions. First, the message aggregation and combination should be simplified to reduce the complexity of calculations.…”

Section: A Principles Of Graph Neural Networkmentioning

confidence: 99%

“…To solve the graph optimization problem effectively, the GNNs which satisfy the permutation equivariance/invariance properties are favorable. To fully reveal the advantage of GNN-based solution for (12), authors in [33] bridged the GNN with a class of COAs, i.e., distributed message passing, to justify that for any graph optimization problem, there exists a GNN that can solve it from the algorithmic perspective. Compared with the COAs, the data driven GNN can achieve near-optimal performance with reduced computational complexity.…”

Section: B Graph Optimization Problems In Wireless Networkmentioning

confidence: 99%

See 4 more Smart Citations

Machine Learning for Large-Scale Optimization in 6G Wireless Networks

Shi¹,

Lian²,

Shi³

et al. 2023

Preprint

View full text Add to dashboard Cite

The sixth generation (6G) wireless systems are envisioned to enable the paradigm shift from "connected things" to "connected intelligence", featured by ultra high density, largescale, dynamic heterogeneity, diversified functional requirements and machine learning capabilities, which leads to a growing need for highly efficient intelligent algorithms. The classic optimization-based algorithms usually require highly precise mathematical model of data links and suffer from poor performance with high computational cost in realistic 6G applications. Based on domain knowledge (e.g., optimization models and theoretical tools), machine learning (ML) stands out as a promising and viable methodology for many complex large-scale optimization problems in 6G, due to its superior performance, generalizability, computational efficiency and robustness. In this paper, we systematically review the most representative "learning to optimize" techniques in diverse domains of 6G wireless networks by identifying the inherent feature of the underlying optimization problem and investigating the specifically designed ML frameworks from the perspective of optimization. In particular, we will cover algorithm unrolling, learning to branch-andbound, graph neural network for structured optimization, deep reinforcement learning for stochastic optimization, as well as endto-end learning for semantic optimization, for solving challenging large-scale optimization problems arising from various important wireless applications. To enable ML implementation in distributed wireless networks across massive number of end devices, federated learning for distributed optimization will further be presented. Through the in-depth discussion, we shed light on the excellent performance of ML-based optimization algorithms with respect to the classical methods, and provide insightful guidance to develop advanced ML techniques in 6G networks. Neural network design, theoretical tools of different ML methods,

show abstract

Section: B Machine Learning In 6gmentioning

confidence: 99%