Sung Eun "Samuel" Kwon, Ph.D.
May 19(Thu) - May 19(Thu), 2022
12PM
Online zoom (ID: 728-142-6028)
Neuro@noon Seminar
Date: 12:00 PM, Thursday, May 19th
Speaker: Sung Eun "Samuel" Kwon, Ph.D.
(Department of Molecular, Cellular, Developmental Biology, University of Michigan)
Title: Learning-related plasticity of population coding in sensory cortex
Abstract: Our ability to categorize external sensory stimuli and initiate an appropriate goal-oriented action is critical for daily lives. Typical responses of cortical neurons to identical sensory stimuli are highly variable and noisy. Much work has investigated the impact of correlated response variability on the amount of information encoded in sensory cortical areas. A dogmatic view is that the correlated variability among similarly tuned neurons limits the amount of sensory information encoded. However, learning-dependent formation of subnetworks – groups of neurons that share elevated correlated variability and similar tuning towards task-associated variables – is frequently observed, even in primary sensory cortices. An emerging view posits that, while subnetworks of highly correlated neurons make sensory encoding less efficient, they benefit decision-making by enhancing the information propagation to downstream areas. Critically, however, it remains relatively unexplored how neuronal correlations change through learning different tasks.
We recorded population activity of layer 2/3 neurons in the whisker somatosensory cortex while head-fixed mice learn to categorize whisker vibration frequency and report by licking the lick port. Additional cohorts of mice were subjected to either repeated whisker stimulation (control) or ‘detection’ task in which animals reported the presence of whisker stimulus without categorizing the stimulus frequency. During category learning, neurons responding to stimuli within the same category display increases in pairwise signal and noise correlations across learning, whereas those in opposite categories show decreased correlations. This result shows that learning drives the formation of categorical ‘subnetworks’ in the L2/3 of wS1. Importantly, removing noise correlations by shuffling trials improves decoding of stimulus category with SVM, suggesting that they are ‘information limiting’. We have tested whether such subnetworks emerge specifically during category learning or ‘simpler’ tasks such as whisker detection.