GenAI-SFC: A GenAI-assisted Deep Reinforcement Learning approach for adaptive SFC provision in 6G intelligent networks

With the advent of the 6G era, Inclusive Intelligent Services (IIS)—which rely on the convergence of advanced communication technologies and Artificial Intelligence (AI) algorithms—are expected to become ubiquitous. These services necessitate efficient distributed intelligent learning and reasoning mechanisms to support diverse and dynamic application scenarios. However, the increasing complexity of emerging applications poses significant challenges for orchestrating differentiated, customized services through efficient Service Function Chaining (SFC) provisioning. SFC refers to the structured arrangement of various Virtual Network Functions (VNFs) to enable seamless service delivery. To address the stringent requirements of 6G IIS, particularly ultra-low latency and high reliability, this study proposes an advanced Deep Reinforcement Learning (DRL) framework augmented with a generative Conditional Variational AutoEncoder (CVAE) for automated and adaptive SFC provisioning. The proposed hybrid approach leverages VAE-driven feature extraction and dimensionality reduction to enhance generalization, while the DRL component optimizes exploration and decision-making through trial-and-error learning. The simulation results demonstrate that the proposed approach significantly improves the performance of 6G intelligent services, surpassing baseline algorithms in key metrics, including provision cost, SFC acceptance rate, and end-to-end SFC delay.