Schedule of Events | Symposia

Predictive power of the divisive normalization model of numerosity perception

Poster Session F - Tuesday, April 1, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Joonkoo Park¹ (joonkoo@umass.edu), Jenna Croteau¹, Michele Fornaciai², David E. Huber³; ¹University of Massachusetts Amherst, ²Université Catholique de Louvain, ³University of Colorado Boulder

Numerosity perception remains a neurocomputational puzzle, as it's unclear how discrete number representations emerge from continuous neural activity. Recently, Park & Huber (2022) proposed a computational model and demonstrated that a simple neural network with center-surround filters and divisive normalization effectively normalizes the dimensions orthogonal to numerosity of an item array, making the neural network sensitive to numerosity. Here, we develop and test two novel predictions that the model makes using electroencephalography. First, the model predicts that coherence illusion for area and orientation (where numerosity perception is influenced by item homogeneity, for example, in area or orientation) arises from two different neural stages. Second, the model predicts that the visual processing stage at which numerosity is first represented contains no information about numerosity if the dot-array images are equalized for spatial frequency amplitude spectrum. To test these predictions, the visual-evoked potentials measured from participants viewing images of item arrays were analyzed to evaluate how early visual cortical activities (80-200 ms) are modulated by the manipulation of item homogeneity (in Study 1) and spatial frequency amplitude spectrum (in Study 2). Study 1 showed that area and orientation coherence modulate visual cortical activities at distinct latencies and topographies, aligning with the model’s prediction. Study 2 showed that equalization of spatial frequency amplitude spectrum abolishes early, but not later, visual cortical sensitivity to numerosity, also in part aligning with the prediction. These findings provide strong empirical support for the divisive normalization model as an algorithmic-level theory of numerosity perception.

Topic Area: PERCEPTION & ACTION: Vision