Population encoding of stimulus features along the visual hierarchy

Abstract

The retina and primary visual cortex (V1) both exhibit diverse neural populations sensitive to diverse visual features. Yet it remains unclear how neural populations in each area partition stimulus space to span these features. One possibility is that neural populations are organized into discrete groups of neurons, with each group signaling a particular constellation of features. Alternatively, neurons could be continuously distributed across feature-encoding space. To distinguish these possibilities, we presented a battery of visual stimuli to mouse retina and V1 while measuring neural responses with multi-electrode arrays. Using machine learning approaches, we developed a manifold embedding technique that captures how neural populations partition feature space and how visual responses correlate with physiological and anatomical properties of individual neurons. We show that retinal populations discretely encode features, while V1 populations provide a more continuous representation. Applying the same analysis approach to convolutional neural networks that model visual processing, we demonstrate that they partition features much more similarly to the retina, indicating they are more like big retinas than little brains.

From Sensory to Perceptual Manifolds: The Twist of Neural Geometry

“…Here in this study, we analyzed neuronal activity data previously recorded in the V2 area 13 , aiming to elucidate how the sensory manifold, where MIC orientations remains linearly inseparable akin to the stimulus space, transforms into the perceptual manifold that affords linear separability. To do this, we utilized the population coding doctrine to characterize neural manifolds constituted by neuron population 14,15,16 , an approach recently applied to various domains such as vision 17,18 , memory 19,20 , decision 21,22,23 , navigation 24,25 and motor execution [26][27][28][29] . Specifically, we first established the correspondence between the sensory manifold and the stimulus space.…”

“…Here in this study, we recorded neuronal activity in the V2 to elucidate how the sensory manifold, where MIC orientations remains linearly inseparable akin to the stimulus space, transforms into the perceptual manifold that affords linear separability. To do this, we utilized the population coding doctrine to characterize neural manifolds constituted by neuron population (SueYeon Chung & Abbott, 2021;Seung HS & Lee DD, 2000;DiCarlo & Cox, 2007), an approach recently applied to various domains such as vision (Froudarakis et al, 2020;Dyballa et al, 2023), memory (Xie et al, 2022;She et al, 2021), decision (Okazawa et al, 2021;Flesch et al, 2022;Genkin et al, 2023), navigation (Gardner et al, 2022;Low et al, 2023) and motor execution (Churchland et al, 2012;Russo et al, 2020;Gallego et al, 2017;Gallego et al, 2018). Specifically, we first established the correspondence between the sensory manifold and the stimulus space.…”