Holistic cell assembly machinery of stimulus representation in visual cortex --why do sensory neurons appear to be noisy?

For a typical neuroscience experiment of repeated stimulus presentations, the observed coordinated neuronal population activity in the cerebral cortex is impacted by both experimenter-controlled stimuli and unobserved variables on each trial.  A fundamental challenge in neuroscience is to infer the unobserved stimulus-related or –unrelated variables from the recorded neural population activity and their across-trial co-variations.  To address this challenge,  we recorded the activity from hundreds of neurons in the visual cortex of awake mice in response to repeated presentation of simple and naturalistic visual stimuli.  We applied an unsupervised low-rank tensor decomposition to the high-dimensional data set.  This approach enabled us to distinguish stimulus-related or –unrelated variables from neuronal population activity.  The approach further revealed co-active groups of neurons (“cell assemblies”) that were stimulus-specific.  In a complementary approach, we determined the single-neuron property of response reliability across trials.  Unexpectedly, unreliable neurons covaried with other neurons to encode stimulus-unrelated unobserved variables. Furthermore, neurons of all levels of response reliability contributed to encode stimulus-related unobserved variables.  These results reveal a holistic cell assembly machinery of stimulus representation in visual cortex in which neurons from the entire range of response reliabilities contribute.