Geometric reshaping of task-relevant representations in the primary visual cortex supports perceptual decisions

Abstract

Discriminating between two stimuli requires that their neural representations become separable by downstream readouts. This can be achieved geometrically, by disentangling the manifolds that population responses form in neural state space. Such reorganization has been observed in associative and motor areas, but never at the earliest stage of cortical processing. While task learning is known to impact neuronal representations in the primary visual cortex (V1), it is unknown if the population geometry is also reshaped to support perceptual decisions. We imaged V1 populations in mice trained on a Go/NoGo orientation discrimination task of increasing difficulty, and in naive mice passively viewing the same stimuli. Training reshaped the representational geometry so that the population responses were better linearly separable. A static compression made the Go and NoGo manifolds more compact and lower-dimensional from the earliest response, while a dynamic separation drove them further apart through the trial. Together, those transformations increased manifold capacity and readout accuracy. This reorganization made the Go-NoGo relationship in the neural state space more stable in Trained animals than in Naive ones. Within this learned geometry, the position of individual trials along the Go-NoGo axis predicted the animals' decision probabilities. Learning therefore promotes a disentangled and stable representational geometry that feeds the decision process.