Using fMRI-Based Multi-Scale Perception Models to Explore Cognitive Load and Attention Allocation in Education

The study of cognitive load and attention allocation has gained prominence in educational research due to its critical role in optimizing teaching strategies. Cognitive load refers to the mental effort required to process information, while attention allocation describes how learners distribute their cognitive resources across tasks. Traditional methods, such as Cognitive Load Theory (CLT) and behavioral measures (e.g., reaction times, eye-tracking), provide valuable insights but fall short in capturing real-time neural mechanisms. These approaches are limited by their reliance on single-task contexts and their inability to probe dynamic, multi-scale brain processes during learning. To address these gaps, we propose a novel framework integrating fMRI-based multi-scale perception models with behavioral and physiological data. This method allows for real-time analysis of brain activity, providing a granular understanding of how cognitive load and attention are distributed across complex learning tasks. Our approach leverages neural data from key brain regions involved in memory and attention, enabling the development of adaptive, personalized educational tools. Technically, our method employs a hybrid CNN-ViT architecture with dynamic uncertainty calibration, enabling high-resolution multimodal fusion across EEG and fMRI signals. Experimental results show that our model achieves up to 91.83% accuracy on the ADHD-200 dataset, significantly outperforming state-of-the-art baselines. Experimental results demonstrate that our method significantly improves the understanding of neural mechanisms underlying learning. By correlating cognitive load with brain oscillations (e.g., theta, alpha bands), we identify biomarkers that predict learning efficiency and attention shifts. This research contributes to the design of adaptive learning systems, paving the way for personalized, scalable educational solutions that align with neural capacities.