Developmental changes in similarity between neural representations of mental arithmetic and artificial neural networks.

The ability to learn simple arithmetic is often attributed to the associative nature of human memory, where repeated exposure strengthens the relations between operands and outcomes. Because artificial neural networks (ANNs) also learn input-output associations, we tested whether ANN-derived features would increasingly capture arithmetic-related brain activity with development. We analyzed fMRI responses to addition problems in 104 participants across four age groups (8-, 11-, 14-year-olds, and adults). An ANN-based encoding model better predicted activity in older than younger participants in the left precentral sulcus. In adults, prediction accuracy was higher for smaller than larger problems. ANN prediction patterns were consistent with increasingly discrete representations of individual addition problems with age. These findings suggest that some neural representations of arithmetic problems become discretely organized with development. However, ANN-derived features captured this change only in the left precentral sulcus, suggesting that associative mechanisms may account for only part of arithmetic processing.