Deciphering the visual cortical code with 10,000-neuron recordings

January 31, 2017 -
12:00pm to 1:00pm
Kenneth Harris; Hosted by Henry Yin

Perception and action emerge from the coordinated activity of large neuronal populations. Many theories exist for the function of these populations, including that cortex recodes sensory inputs into a sparse, efficient format; and that spontaneous activity recapitulates patterns seen during sensory processing, reflecting the brain's expectations for potential sensory inputs. While recordings of up to hundreds of neurons have provided glimpses of the neuronal population code, these are not sufficient to reveal the high-dimensional structure of neural coordination.

We used 2-photon calcium imaging and improved analysis methods to record from populations of >10,000 neurons in the visual cortex of awake mice, in response to thousands of natural images. We found that nearly all neurons were sparsely tuned, and no two neurons had similar stimulus tuning. Stimulus correlations were significant in only ~1% of neuron pairs, and population responses spanned the full-dimensional space of stimuli. These results indicate sparse, efficient coding of natural images.

Spontaneous activity was prominent in our recordings, spanning ~100 dimensions of shared population activity, and trial-to-trial "noise" in sensory responses could be largely explained by addition of these spontaneous dimensions onto fixed stimulus responses.