A Low-latency Adaptive Learning System Model Supported by EdgeComputing

Introduction: Adaptive learning systems face a fundamental tension between edge device resource constraints and high-precision demands. Methods: The study proposed a four-level terminal-edge-fog-cloud collaborative architecture integrating K-means geographic clustering, temperature-guided knowledge distillation from ResNet-50 to MobileNetV3, CPU-load-triggered dynamic pruning, and an entropy-driven EDF scheduler fused with incremental Deep Q-Network (DQN) reinforcement learning. results: Experiments on the EdNet 2021-2023 dataset, based on a 500-node Raspberry Pi cluster, demonstrate that the model achieved an average end-to-end latency of only 99ms under high load, with a recommendation accuracy of 95%, significantly outperforming existing baseline methods. Results: Experiments on a 500-node Raspberry Pi cluster using the EdNet 2021–2023 dataset show an average end-to-end latency of 99 ms under high load with 95% recommendation accuracy. CPU utilization remains below 70%, and student learning gains improved by 22.6% compared with 15.5% in the control group. Discussion: The framework resolves the “compute-accuracy” trade-off by co-designing spatial deployment, model compression, real-time adaptation, and latency-aware scheduling. Conclusion: This work establishes a replicable, low-latency, high-accuracy adaptive learning paradigm for resource-constrained edge environments.