Limits and Distribution of Information in Large Neural Populations


August 4, 2020 - 1:00pm - 2:30pm
About the Speaker
Speaker Title: 
Jan Drugowitsch
Speaker Affiliation: 
Assistant Professor of Neurobiology, Harvard Medical School

Neither humans nor animals are perfect at detecting minute differences
in our sensory experiences. Somewhere within our brains information
about these differences is limited. What is the origin of this
limitation? Does it emerge from the noisy biological units that make
up our brains? Or is the input information already limited? We
distinguished these alternatives by measuring how information about a
visual stimulus scales with the number of neurons in a mouse V1
population. This scaling revealed that limited input information
rather than noise neurons is the dominant reason for limited
information. Furthermore, we found that information is distributed
homogeneously across the whole neural population. As a consequence, it
is sufficient to sample random subpopulation in order to recover most
information. Our findings suggest that brain performance isn't limited
by the intrinsic noise of its biological units, but instead by limited
input information and the complexity of the algorithms it implements.