Attention Decorrelates Sensory and Motor Signals in the Mouse Visual Cortex

Abstract

Visually-guided behaviors depend on the activity of cortical networks receiving visual inputs and transforming these signals to guide appropriate actions. However, non-retinal inputs, carrying motor signals as well as cognitive and attentional modulatory signals, also activate these cortical regions. How these networks avoid interference between coincident signals ensuring reliable visual behaviors is poorly understood. Here, we observed neural responses in the dorsal-parietal cortex of mice during a visual discrimination task driven by visual stimuli and movements. We found that visual and motor signals interacted according to two canonical mechanisms: divisive normalization and response demixing. Interactions were contextually modulated by the animals state of attention, with attention amplifying visual and motor signals and decorrelating them in a low-dimensional space of neural activations. These findings reveal canonical computational principles operating in dorsal-parietal networks that enable separation of incoming signals for reliable visually-guided behaviors during interactions with the environment.