Mixture of Experts for Network Optimization: A Large Language Model-enabled Approach

Optimizing various wireless user tasks poses a significant challenge fornetworking systems because of the expanding range of user requirements. Despiteadvancements in Deep Reinforcement Learning (DRL), the need for customizedoptimization tasks for individual users complicates developing and applyingnumerous DRL models, leading to substantial computation resource and energyconsumption and can lead to inconsistent outcomes. To address this issue, wepropose a novel approach utilizing a Mixture of Experts (MoE) framework,augmented with Large Language Models (LLMs), to analyze user objectives andconstraints effectively, select specialized DRL experts, and weigh eachdecision from the participating experts. Specifically, we develop a gatenetwork to oversee the expert models, allowing a collective of experts totackle a wide array of new tasks. Furthermore, we innovatively substitute thetraditional gate network with an LLM, leveraging its advanced reasoningcapabilities to manage expert model selection for joint decisions. Our proposedmethod reduces the need to train new DRL models for each unique optimizationproblem, decreasing energy consumption and AI model implementation costs. TheLLM-enabled MoE approach is validated through a general maze navigation taskand a specific network service provider utility maximization task,demonstrating its effectiveness and practical applicability in optimizingcomplex networking systems.