Comparison of visual quantities in untrained neural networks

Area of Science:

• Neuroscience
• Computational Neuroscience
• Developmental Neuroscience

Background:

• The capacity for comparing visual quantities, using both proportion and difference, is present from birth in animals.
• The neural mechanisms underlying this innate ability, particularly before visual experience, are not well understood.

Purpose of the Study:

• To propose and investigate a computational model for the spontaneous emergence of neuronal tuning for visual quantity comparisons in untrained neural networks.
• To explore how distinct comparison measures (proportion vs. difference) might arise from fundamental neural properties.

Main Methods:

• Development of a biologically inspired model neural network with random initializations.
• Analysis of neuronal activity within the network to identify units selective for proportion and difference.
• Simulation of quantity comparison tasks using the developed network.

Main Results:

• Spontaneous emergence of single units selective for proportions and differences in randomly initialized feedforward networks.
• Demonstration that these emergent units enable the network to perform visual quantity comparison tasks.
• Identification that distinct tunings for proportion and difference arise from the summation of nonlinear neural activities, with subtle differences in response functions determining the measure type.

Conclusions:

• Visual quantity comparison, utilizing both proportion and difference, represents a primitive brain function.
• These abilities can emerge spontaneously in neural circuits prior to extensive learning or visual experience.
• The findings suggest a foundational neural basis for quantity perception present early in development.