A Multi-Agent Reinforcement Learning and Diffusion Model Hybrid for Real-Time Personalized Curriculum Optimization in Intelligent Education Platforms

Abstract

Adaptive learning systems are critical for achieving personalized curriculum optimization on intelligent education platforms, which must accommodate the varied behaviors of learners, evolving areas of knowledge, and instantaneous educational decisions. Traditional recommendation systems have difficulty generating dynamic, situation-dependent learning paths, as traditional education paradigms are being continuously changed by learning events, behaviors, and conditions. To address these issues, a hybrid approach is proposed that uses Multi-Agent Reinforcement Learning (MARL) together with a diffusion-based model to find real-time, personalized curriculum optimization. With this approach, a multi-agent cooperative reinforcement learning system is presented to manage learner profiles, curriculum sequencing, assessments, and resources, while the diffusion model provides curriculum optimization through denoising processes. This hybrid framework continuously updates educational pathways through aggregating learner interaction histories, engagement trends, and reward-optimized performance. In the development of the proposed system, please note that the datasets used for evaluation were both large and used for intelligent tutoring systems, were made up of over 120,000 instances of learner interactions based on the learner, and were used as a comparison against various different approaches (Deep Knowledge Tracing, transformer-based recommendation, and Reinforcement Learning methods). Experimental results indicate that the proposed framework achieved a Curriculum Adaptation Score of 96.8, Learning Engagement Gain of 21.4%, Knowledge Retention Index of 93.7%, Curriculum Stability Factor of 0.91, and Recommendation Latency Reduction of 24.9%. An ablation analysis of the system has further shown that the coupling of MARL coordination and diffusion of the curricula produced a significant improvement in adaptive educational recommendation and learner-specific curriculum personalization. As such, these results demonstrate that the proposed MARL-Diffusion framework provides an efficient and scalable method for intelligent education platforms through the improvement of adaptive curriculum sequencing, learner optimization in terms of engagement, and efficiency of real-time educational recommendation performances during dynamic learning environments.