Active filtering of sequences of neural activity by recurrent circuits of sensory cortex

Authors

• CD Deveau
• Z Zhou
• PK LaFosse
• Y Deng
• S Mirbagheri
• N Steinmetz
• MH Histed

Abstract

In daily life, organisms interact with a sensory world that dynamically changes from moment to moment. Recurrent neural networks can generate dynamics, but in sensory cortex any dynamic role for the dense recurrent excitatory-excitatory network has been unclear. In this work we show a new role for recurrent connections in mouse visual cortex: they support powerful dynamical computations, but via filtering sequences of input, instead of generating sequences. Using two-photon optogenetics, we measure responses to natural images and play them back, showing amplification when played back during the correct movie dynamic context and suppression in the incorrect context. The sequence selectivity depends on a network mechanism: inputs to groups of cells produce responses in different local neurons, which interact with and change responses to later inputs. We confirm this mechanism by designing sequences of inputs that are amplified or suppressed by the network. Finally, we show the experimental data matches predictions from artificial recurrent networks, as recurrent networks trained to amplify sequences show the interactions from one pattern to the next that we find experimentally. Together, these results suggest a novel function, sequence filtering, for recurrent connections in the cerebral cortex. This implies that the many recurrent excitatory-excitatory connections in the sensory cortex learn via development and experience the statistics of the natural world, encoding this information in recurrent synaptic weights.

Scientific Focus Area: Neuroscience

This page was last updated on Tuesday, August 6, 2024