Emergence of behavioral phenomena and adaptation effects in human numerosity decoder using recurrent neural networks

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Mathematical Psychology

Background:

• Humans possess an innate ability for visual numerosity perception, the cardinality of a set.
• The lateral intraparietal cortex (LIP) and intraparietal sulcus (IPS) are key brain regions for number processing.
• Existing computational models of number perception have limitations in range and behavioral characteristic simulation.

Purpose of the Study:

• To develop a novel computational model for number perception.
• To account for a wider range of numerosities and behavioral characteristics like adaptation effects.
• To elucidate the neural mechanisms underlying numerosity perception.

Main Methods:

• Utilized a neural network with self-excitatory and mutual inhibitory properties.
• Assumed mean network activation at steady state encodes numerosity with a monotonically increasing relationship to set size.
• Optimized inhibition strengths to cover distinct numerical ranges (1-4, 5-17, 21-50).

Main Results:

• The model successfully encodes numerosities across three distinct intervals, aligning with behavioral breakpoints.
• Achieved decoding of mean activation into a continuous scale from 1 to 50.
• Proposed a dynamic inhibition strength selection mechanism for extended numerosity range operation.

Conclusions:

• The novel computational model provides insights into the neural substrates of number perception.
• The model explains diverse behavioral phenomena, including adaptation effects and continuous visual attribute influences.
• This work advances our understanding of how the brain processes numerical information.