A Multiscale Transformer Framework for Optimizing Educational Resource Transmission in Preschool Wireless Networks

Abstract

This paper proposes EduTransNet, a novel framework combining cross-scale window attention mechanisms with separable spatio-temporal attention to enhance network transmission efficiency and resource utilization in preschool wireless networks. The framework jointly models temporal dependencies and long-range network node relationships, incorporating multiresolution optimization strategies for adaptive resource allocation. Experimental results on the EdNet dataset, containing over 131 million student interactions, demonstrate that EduTransNet achieves significant improvements with a PSNR of 37.13 dB and SSIM of 0.978, surpassing existing methods by 2.3 dB and 0.008, respectively. The framework shows particular strength in handling dynamic educational content delivery scenarios with multiple concurrent users while maintaining a low latency of 160 ms.