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

TL;DR

This survey investigates how generative AI (GenAI)—specifically generative diffusion models (GDMs) and large pre-trained models (LPTMs)—applies to network optimization under one-shot and Markov decision process (MDP) formulations. It provides a structured taxonomy, surveys representative work, and derives theoretical generalization bounds for GDM-based generation and policy decisions, while critically examining limitations such as constraint satisfaction and probabilistic outputs. The authors argue for bridging generation and optimization through modular, rule-guided generation and collaboration with traditional solvers and humans, complemented by virtualization for safe evolution. The work offers practical guidance for deploying GenAI in 6G-era networks and outlines concrete future directions, benchmarks, and datasets to advance theory and practice in this domain.

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.

Figures (7)

• Figure 1: GenAI in network optimization : taxonomy map.
• Figure 2: Roadmap of this survey.
• Figure 3: The two modeling paradigms for network optimization and the location in network systems.
• Figure 4: Comparison between GDMs and LLMs in terms of generation mechanism. The sources of randomness in GDMs and LLMs differ: in GDMs, randomness arises from the initial noise sample and the stochastic nature of the denoising trajectory, whereas in LLMs, it stems from sampling predicted tokens. Moreover, LLMs are autoregressive, as each predicted token is fed back into the model for subsequent predictions, while most GDMs are non-autoregressive.
• Figure 5: Comparison between BBO and solution generation in the networking domain. Here, $\mathbf{\hat{y_i^*}}$ denotes the hypothetical optimal solution for input $\mathbf{x}_i$, while $\mathbf{\tilde{y_i^*}}$ represents the corresponding generated solution.

Theorems & Definitions (1)

• Remark : Generalization Bound