A canonical neural circuit for motion direction estimation

Abstract Abundant experimental data support the view that motion detection is a canonical neural operation, common to a wide variety of species and different stages in the visual stream. However, existing motion models lack this generality, as their predictive power tends to be restricted to some specific visual phenomena, brain regions, or animal taxa. Here, we introduce a neural circuit model that only uses cells selective to cardinal directions, with elements that can be interpreted as emulating some dendritic processes, and we show how it is able to predict a wide range of visual motion data, from basic properties to hitherto unexplained phenomena, surpassing state-of-the-art models implemented as deep neural networks with millions of parameters. The direction-selective cells that our model relies on are present in the visual systems of many classes of animals, from insects to primates, and in visual pathway structures going from the retina to the cortex, suggesting that the proposed model could serve as a canonical neural circuit for estimating motion direction.