Unifying network model links recency and central tendency biases in working memory.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Behavioral Neuroscience

Background:

• Central tendency bias (contraction bias) influences working memory judgments toward the average of past observations.
• Recency biases, like serial dependence, are thought to reflect current working memory content.
• These biases were traditionally viewed as distinct phenomena.

Purpose of the Study:

• To investigate the relationship between central tendency bias and recency biases in working memory.
• To explore the neural mechanisms underlying these biases, particularly the role of the posterior parietal cortex (PPC).
• To develop and validate a computational model explaining the generation of these biases.

Main Methods:

• Auditory delayed comparison task in rats.
• Silencing of the posterior parietal cortex (PPC) in rats.
• Development of a computational model of working memory circuits.
• Human psychophysics experiments using an auditory parametric working memory task.

Main Results:

• Silencing the PPC in rats significantly reduced both short-term biases and contraction bias.
• A computational model demonstrated that volatile working memory susceptible to sensory history naturally produces contraction bias.
• Model errors are sampled from the full stimulus distribution, not a gradual shift towards the mean.
• Human psychophysics experiments validated the model's predictions across various parameters.

Conclusions:

• Central tendency bias and short-term sensory history biases may arise from a unified neural mechanism involving volatile working memory.
• The posterior parietal cortex plays a crucial role in mediating these biases.
• The proposed model provides a parsimonious explanation for a range of behavioral findings in working memory.