Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning State-Augmented Policies for Information Routing in Communication Networks

Sourajit Das, Navid NaderiAlizadeh, Alejandro Ribeiro

TL;DR

The paper addresses routing in large-scale communication networks under constraints by formulating it as a constrained learning problem using only local information. It introduces State Augmentation (SA) with Graph Neural Networks (GNNs) to parameterize routing policies, leveraging the Augmented Lagrangian (MoM) framework to achieve fast, feasible, near-optimal solutions while enabling distributed implementation. The proposed method demonstrates strong performance in terms of utility and queue stability, and shows robustness to graph perturbations and transferability to unseen graph sizes and real topology graphs. The work highlights the practicality of deploying SA-GNN policies for scalable, decentralized information routing with convergence guarantees and applicability to real-world networks.

Abstract

This paper examines the problem of information routing in a large-scale communication network, which can be formulated as a constrained statistical learning problem having access to only local information. We delineate a novel State Augmentation (SA) strategy to maximize the aggregate information at source nodes using graph neural network (GNN) architectures, by deploying graph convolutions over the topological links of the communication network. The proposed technique leverages only the local information available at each node and efficiently routes desired information to the destination nodes. We leverage an unsupervised learning procedure to convert the output of the GNN architecture to optimal information routing strategies. In the experiments, we perform the evaluation on real-time network topologies to validate our algorithms. Numerical simulations depict the improved performance of the proposed method in training a GNN parameterization as compared to baseline algorithms.

Learning State-Augmented Policies for Information Routing in Communication Networks

TL;DR

The paper addresses routing in large-scale communication networks under constraints by formulating it as a constrained learning problem using only local information. It introduces State Augmentation (SA) with Graph Neural Networks (GNNs) to parameterize routing policies, leveraging the Augmented Lagrangian (MoM) framework to achieve fast, feasible, near-optimal solutions while enabling distributed implementation. The proposed method demonstrates strong performance in terms of utility and queue stability, and shows robustness to graph perturbations and transferability to unseen graph sizes and real topology graphs. The work highlights the practicality of deploying SA-GNN policies for scalable, decentralized information routing with convergence guarantees and applicability to real-world networks.

Abstract

This paper examines the problem of information routing in a large-scale communication network, which can be formulated as a constrained statistical learning problem having access to only local information. We delineate a novel State Augmentation (SA) strategy to maximize the aggregate information at source nodes using graph neural network (GNN) architectures, by deploying graph convolutions over the topological links of the communication network. The proposed technique leverages only the local information available at each node and efficiently routes desired information to the destination nodes. We leverage an unsupervised learning procedure to convert the output of the GNN architecture to optimal information routing strategies. In the experiments, we perform the evaluation on real-time network topologies to validate our algorithms. Numerical simulations depict the improved performance of the proposed method in training a GNN parameterization as compared to baseline algorithms.
Paper Structure (20 sections, 41 equations, 11 figures)

This paper contains 20 sections, 41 equations, 11 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Statement
  4. Problem Formulation
  5. Solving for Gradient Based Routing in the Lagrangian Domain
  6. Solution using the Augmented Lagrangian: Method of Multipliers
  7. The Alternating Direction Method of Multipliers (ADMM)
  8. Graph Neural Networks based Parameterization for Routing Optimization
  9. GNN Parameterization with Augmented Lagrangian: Method of Multipliers
  10. The State Augmentation Algorithm
  11. Implementation Scenario and Practical Assumptions
  12. Experimental Observations & Numerical Results
  13. Communication Network Architectures
  14. Performance Across Different Unparameterized Algorithms
  15. State-Augmentation vs ADMM
  16. ...and 5 more sections

Figures (11)

  • Figure 1: Comparison of two different unparameterized methods of MoM and Dual Descent (DD) for a network with 10 nodes and 5 flows, where the network is run only for a single time step.
  • Figure 2: Performance comparison between unparameterized ADMM and the proposed parameterized state-augmented method using GNNs for networks with 10 nodes and 5 flows, run over $T=100$ time steps.
  • Figure 3: Performance comparison of state augmentation and ADMM algorithms for networks with 5 flows and $N\in\{10, 50, 100\}$ nodes.
  • Figure 4: Performance of state augmentation and ADMM algorithms for network with 50 nodes and $K\in\{5, 10, 15\}$ flows.
  • Figure 5: Relative performance of state-augmented algorithm with ADMM for a network with 50 nodes and 5 flows.
  • ...and 6 more figures