Noise in neural populations accounts for errors in working memory

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Neurobiology

Background:

• Short-term memory errors increase with information quantity, limiting cognitive complexity.
• Previous models of visual memory errors focused on resource allocation, lacking a clear biological basis.
• Alternative theories propose neural noise in neuronal populations as a cause for recall errors.

Purpose of the Study:

• To investigate the biological basis of working memory errors using a neural noise perspective.
• To model human recall failures under increasing memory load using probabilistic spiking neurons.
• To explain deviations from normal distributions in working memory errors and the prioritization of memory representations.

Main Methods:

• Simulated probabilistic spiking neurons with decreasing signal strength to model memory load.
• Analyzed deviations from normal distributions in neural activity to explain recall error patterns.
• Investigated how input drive to neuronal populations affects memory precision for prioritized stimuli.

Main Results:

• Neural noise in probabilistically spiking neurons accurately reproduced human recall error patterns under increasing memory load.
• Deviations from normal distributions, previously attributed to guesses, naturally arise from decoding tuned neuronal populations.
• Altering neuronal input drive demonstrated how memory representations are prioritized for behaviorally relevant information, impacting precision.

Conclusions:

• Neural noise provides a viable biological mechanism for understanding working memory limitations and errors.
• The model explains complex recall error patterns and the adaptive prioritization of memory representations.
• Human observers optimally utilize predictive cues within the constraints of neural noise, demonstrating efficient memory management.