Efficient Coding in Visual Working Memory Accounts for Stimulus-Specific Variations in Recall.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Working memory recall errors exhibit both bias and precision that depend on stimulus properties.
• Existing models struggle to explain stimulus-specific variations in recall error, often relying on ad hoc explanations.

Purpose of the Study:

• To develop a principled neural model of working memory that accounts for stimulus-specific recall variations.
• To investigate the role of efficient coding and environmental statistics in shaping visual working memory performance.

Main Methods:

• Adapted a neural model of working memory incorporating the principle of efficient coding.
• Incorporated an orientation prior favoring cardinal over oblique orientations based on natural stimulus frequencies.
• Tested the model's predictions in human subjects by manipulating the distribution of nontarget stimuli during adaptation blocks.

Main Results:

• The model accurately captured stimulus-specific variations in recall error, outperforming a slot-based model.
• Efficient coding predicted a repulsive bias away from cardinal orientations.
• Exposure to incongruent stimulus distributions (peaks at oblique orientations) reduced this repulsion in human subjects, persisting post-adaptation.

Conclusions:

• The unified neural framework provides a principled account of stimulus-specific effects in working memory.
• Environmental statistics and prior expectations significantly influence visual working memory biases.
• The findings validate the neural framework and highlight the importance of efficient coding in orientation representation.