A DeepSeek cross-modal platform for personalized art education in Autism Spectrum Disorder

Abstract Educational inequity in arts learning disproportionately marginalizes students with Autism Spectrum Disorder (ASD), who require structured, predictable environments for aesthetic development and sensory regulation that traditional pedagogies fail to provide. This study introduces an AI-powered e-learning platform that addresses these systematic barriers through intelligent cross-modal integration, democratizing access to personalized art education for neurodivergent learners. Our DeepSeek-based system transforms visual art features into structured musical accompaniments that accommodate individual sensory processing patterns, cultural backgrounds, and neurodevelopmental profiles while maintaining the predictability essential for ASD learning success. The platform employs enhanced ResNet-50 architecture, high-dimensional manifold mapping, and conditional generation models specifically optimized for sensory regulation principles. Comprehensive evaluation with 203 participants (including 53 neurodivergent learners) and 19 autism education specialists demonstrates substantial improvements: sensory comfort ratings of 4.6/5, learning satisfaction of 4.3/5, and educational outcomes showing 20.5% NAEP score improvements compared to 8.2% for traditional methods (p < 0.008). Technical performance achieved superior cross-modal consistency (MSE 0.05, PCC 0.92) with 89% accommodation success across diverse sensory profiles. This research offers a promising model for inclusive digital education by demonstrating how AI can contribute to mitigating educational inequities for neurodivergent populations. It provides a scalable framework that advances accessible arts education, embracing neurodiversity while maintaining academic rigor.