Growing Minds, Integrating Senses: Neural and Computational Insights into Age-related Changes in Audio-Visual and Tactile-Visual Learning in Children

Abstract

Multisensory processing and learning shape cognitive and language development, influencing how we perceive and interact with the world from an early age. While multisensory processes mature into adolescence, it remains poorly understood how age influences multisensory associative learning. This study investigated age-related effects on multisensory processing and learning during audio-visual and tactile-visual learning in 67 children (5.7-13 years) by integrating behavioural and neuroimaging data with computational methods. A reinforcement-learning drift diffusion model revealed that older children processed information faster and made more efficient decisions on multisensory associations. These age-related increases coincided with higher activity in brain regions associated with cognitive control, multisensory integration, and memory retrieval, specifically during audio-visual learning. Notably, the bilateral anterior insula exhibited heightened activation in response to lower reward prediction errors, indicative of increased sensitivity to negative feedback with development. Finally, reward prediction errors modulated activation in reward processing and cognitive control regions, with this modulation remaining modality-independent and largely stable across age. In conclusion, while children employ similar learning strategies, older children make decisions more efficiently and engage neural resources more strongly. Our findings reflect ongoing maturation of neural networks supporting multisensory learning in middle childhood, enabling more adaptive learning in later childhood. Highlights- faster information processing in older children during a multisensory learning task - increasing brain activation with age in visual, parietal, and frontal regions - reward prediction error processing are independent of age and modality - heightened response to negative reward prediction errors with increasing age