Frontiers of Generative AI for Network Optimization: Theories, Limits, and Visions

TL;DR

This survey critically reviews the application of generative AI in network optimization, analyzing dominant models, theoretical bounds, limitations, and future directions to guide research in this emerging field.

Contribution

It introduces a categorization of network optimization problems, reviews frontier applications, and provides theoretical insights into generative diffusion models and large pre-trained models.

Findings

Identifies key limitations such as constraint satisfaction and concept understanding.

Provides theoretical generalization bounds for GDMs in different settings.

Highlights the gap between generation and optimization objectives.

Abstract

While interest in the application of generative AI (GenAI) in network optimization has surged in recent years, its rapid progress has often overshadowed critical limitations intrinsic to generative models that remain insufficiently examined in existing literature. This survey provides a comprehensive review and critical analysis of GenAI in network optimization. We focus on the two dominant paradigms of GenAI including generative diffusion models (GDMs) and large pre-trained models (LPTMs), and organize our discussion around a categorization we introduce, dividing network optimization problems into two primary formulations: one-shot optimization and Markov decision process (MDP). We first trace key works, including foundational contributions from the AI community, and categorize current efforts in network optimization. We also review frontier applications of GDMs and LPTMs in other networking tasks, providing additional context. Furthermore, we present theoretical generalization bounds for GDMs in both one-shot and MDP settings, offering insights into the fundamental factors affecting model performance. Most importantly, we reflect on the overestimated perception of GenAI's general capabilities and caution against the all-in-one illusion it may convey. We highlight critical limitations, including difficulties in constraint satisfying, limited concept understanding, and the inherent probabilistic nature of outputs. We also propose key future directions, such as bridging the gap between generation and optimization. Although they are increasingly integrated in implementations, they differ fundamentally in both objectives and underlying mechanisms, necessitating a deeper understanding of their theoretical connections. Ultimately, this survey aims to provide a structured overview and a deeper insight into the strengths, limitations, and potential of GenAI in network optimization.