Abstract:
Examining visual world and calculating the corresponding ensemble measures or perceptual averaging, is vital in ensuring a unitary perceptual world in cognitive agents. Although applying filters to these scenes may seem straightforward, the real difficulty lies in the brain’s ability to fluidly transition between ensemble processing and more attentionally demanding process of individuation across different reference frames. The present work investigates how a neural model can flexibly alternate between these two processes. We utilize a fully connected recurrent neural network with self-excitation and lateral inhibition, which has been used in previous studies for tasks related to enumeration and individuation, to demonstrate its capacity for extracting summary statistics through two separate measures. The results not only confirm the viability of the model, but also offer valuable predictions into ensemble processing in the brain along with possible time scales needed.