The brain makes use of canonical computations that are performed across species, modalities and cortical regions. Determining these canonical computations is crucial in a deep understanding of the brain function. One such computation is normalization, based on which the response of a neuron to a stimulus is dampened by the activity of the neighbouring neural population. Normalization was first proposed for neurons in the primary visual cortex of cat in 1992, and it has since been reported in the visual, auditory, and olfactory systems of different species. However, in the human brain, evidence for its existence is scarce, especially in the presence of attention.

In this work, we used human fMRI data to investigate the role of normalization in the human brain. We compared the predictions of the normalization model with those of the weighted sum model, the weighted average model, and a nonlinear model based on weighted average and response saturation. Our results demonstrated that only the normalization model can predict cortical responses and the observed effects of attention on the response. We thus showed that the normalization model can predict responses to isolated and cluttered stimuli in the presence and absence of attention, providing compelling evidence for its role in the human brain.