KAIST Research Team,
Explains the Principle of Spontaneous Formation of Innate Numerical Comparison Ability
A study has found that the ability of humans and animals to perceive and compare numbers is innately endowed by the structure of neural networks in the brain, rather than acquired through learning.
The Korea Advanced Institute of Science and Technology (KAIST) announced on the 7th that a research team led by Professor Baek Se-bum from the Department of Brain and Cognitive Sciences explained the principle by which innate quantity comparison abilities found in the brain are spontaneously formed.
The function of comparing the quantities of given objects is essential for the survival of animals and humans. In many situations such as conflicts between animal groups, hunting, and food gathering, animals’ decision-making and behavior must vary according to the ratio or difference in quantities of given variables. Observations of the behavior of young individuals who have not undergone learning suggested that quantity comparison ability is an innate brain function, but an explanation of the principle by which this ability arises without learning had not yet been provided.
The research team used brain-inspired artificial neural network models to prove that the cognitive function of perceiving visual quantity ratios and differences can spontaneously emerge in a deep neural network structure without any learning. They also explained that the function of comparing two different types of information?ratio and difference of quantities?can be derived from a common underlying principle.
First, the team discovered that individual neurons selectively responding to the ratio and difference of two quantities spontaneously emerged in a neural network that had undergone no learning at all. When visual quantity information with various ratios or differences was given to an initialized deep neural network, many neurons selectively responding to these were found. Furthermore, the neural activity measured from these neurons was confirmed to be very similar to the neural activity characteristics observed in actual animal experiments. Using this, the team confirmed that they could reproduce a significant portion of the quantity comparison behavioral characteristics reported in animals so far.
Additionally, the research team explained and verified the developmental principle of the neural circuit structure for quantity comparison function through a computational neuroscience model. By analyzing the characteristic connectivity of ratio/difference-selective neurons found in the neural network, they showed that selectivity for specific values can be formed through the combination of simple increasing and decreasing neural activities that spontaneously emerge in lower layers of the neural network. They also confirmed that depending on the type of nonlinearity observed when these neural activities increase or decrease, neurons can differentiate into those perceiving quantity ratio or quantity difference, respectively.
Professor Baek said, “This study showed that the brain’s quantity cognition, comparison, and calculation functions, which were thought to require a considerable amount of learning, can spontaneously emerge from the structure of the initial brain without any learning. By suggesting that various innate higher cognitive functions can arise from the structural and physical properties of the early neural network, this research can provide meaningful directions not only for brain neuroscience but also for new conceptual artificial intelligence research.”
This study was published on the 29th of last month in ‘Cell Reports,’ the online sister journal of the international academic journal ‘Cell.’
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
