ABSTRACT The deployment of large language models in wireless educational environments faces significant challenges due to the tension between computational intensity and dynamic network constraints. This paper proposes a reinforcement learning‐based adaptive inference scheduling framework for Chinese language education systems that intelligently balances service quality and system efficiency under varying wireless conditions. We formulate the scheduling problem as a Markov Decision Process and employ a Double Deep Q‐Network algorithm to learn optimal policies for selecting between full model inference, lightweight model execution, and cached response strategies. Our framework incorporates a comprehensive Quality of Service assessment model. Experimental evaluation across four network conditions demonstrates superior performance. The proposed approach achieves strategy selection accuracy of 89.4%–95.8% while maintaining stable educational quality across dynamic network environments, providing an effective solution for deploying AI‐driven educational systems in resource‐constrained wireless networks.
Wei Li (2026) studied this question.