Population response magnitude variation in inferotemporal cortex predicts image memorability

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Inferotemporal cortex (IT) primarily studied for distinct neural spike patterns in image encoding.
• Limited understanding of how image memorability is modulated by neural population responses.

Purpose of the Study:

• Investigate the role of neural population response magnitude in visual memory.
• Explore the origins of image memorability modulation in the inferotemporal cortex.
• Connect findings from biological systems to artificial neural networks.

Main Methods:

• Analyzing neural data from inferotemporal cortex in monkeys during a visual memory task.
• Probing convolutional neural network models trained for object categorization.
• Correlating population response magnitudes with image memorability in both biological and artificial systems.

Main Results:

• Population response magnitude, not just spike patterns, significantly accounts for image memorability.
• Natural images evoke varying magnitude responses in both the brain and convolutional neural networks.
• Higher layers of neural networks show larger magnitude responses correlated with higher memorability for humans and monkeys.

Conclusions:

• Variation in inferotemporal cortex population response magnitude is a natural outcome of optimized visual processing.
• Population response magnitude plays a crucial role in visual memory formation and recall.
• Convolutional neural networks mirror biological systems in linking response magnitude to memorability.