Stochastic sampling provides a unifying account of visual working memory limits

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Working memory research is characterized by competing models of continuous versus discrete internal representations.
• Discrepancies exist between existing models, hindering a unified understanding of working memory limitations.

Purpose of the Study:

• To introduce a novel sampling approach, grounded in neural coding principles, as a unifying framework for working memory.
• To reconcile seemingly opposing models of working memory by re-conceptualizing them within the sampling framework.
• To identify the critical factors underlying human working memory performance.

Main Methods:

• Developed a computational framework based on sampling principles from neural coding.
• Re-analyzed existing working memory models through the lens of the sampling approach.
• Investigated the role of sampling variability in reproducing human behavioral data.

Main Results:

• The sampling framework reveals commonalities and specific differences between competing working memory models.
• The discrete versus continuous nature of sampling is less critical than random variability in sample counts for model accuracy.
• Stochastic sampling naturally leads to a probabilistic limit on retrieval, without explicit enforcement mechanisms.

Conclusions:

• Established a unified computational framework for working memory, compatible with neural principles.
• Resolved discrepancies between previous theoretical accounts of working memory.
• Highlighted the importance of random variability in sample counts for explaining working memory performance.