Kyoto and Harvard team: noise in mouse visual cortex only slows information growth, not caps it
Recordings of 18–21 thousand visual cortex neurons in five mice showed that shared noise does not cap the information the population encodes, only slows its growth.
Neuroscientists from Kyoto University, Harvard University and partner institutions have studied why the brain can process large volumes of data without failing, Day.Az reports. According to the authors, neural "noise", the random fluctuations in the activity of cells, does not prevent the brain from perceiving more and more information. It only slows that process slightly day.az
Day.Aznews.day.azOpen source ↗. The study was published in Science Advances.
Recordings from the mouse visual cortex
The work was done on the primary visual cortex of mice. The researchers analyzed the activity of 18 to 21 thousand neurons in five rodents that were shown visual stimuli with barely noticeable differences. They measured how groups of cells of different sizes respond to these changes. The authors report that even when the signals of many neurons fluctuate in sync and create interference, the brain keeps extracting new information about the surrounding world without losing efficiency.
To identify the mechanism, the team repeatedly selected groups of neurons at random and split their noise into separate activity patterns. This is how they found two mathematical scaling laws that hold regardless of the size of the cell population.
Two scaling laws and the 30-year assumption
One law describes the strength of the interference. The other describes how each noise pattern combines with the useful signal day.az
Day.Aznews.day.azOpen source ↗. These regularities allowed the researchers to predict the growth of information for groups of neurons that were not observed directly.
"For 30 years everyone expected that common fluctuations of neural activity would inevitably lead to saturation and limit the amount of transmitted information," said lead author Hideaki Shimazaki. "However, our results convincingly show that this limit is not necessary".
A single neuron is unstable: its response to the same image changes from trial to trial. To compensate, the brain uses population coding, sending the same signal through thousands of cells. For a long time it was believed that synchronous noise sets a hard limit on this strategy. According to the study, the brain instead distributes the useful signal over less noisy parts of the nervous tissue. The finding extends the general theory of information scaling.
