Cross-fixation interactions of orientations suggest high-to-low-level decoding in visual working memory

Area of Science:

• Neuroscience
• Cognitive Science
• Visual Perception

Background:

• Sensory encoding progresses from low- to high-level features.
• Visual decoding is assumed to follow a similar low-to-high hierarchy.
• Traditional models posit orientation decoding occurs in low-level visual areas (e.g., V1) without cross-fixation interactions.

Purpose of the Study:

• To investigate the hierarchy of visual decoding in working memory.
• To determine if task-relevant orientations interact across visual fields.
• To challenge the assumption of a strict low-to-high-level decoding hierarchy.

Main Methods:

• Experimental manipulation of visual stimuli presented across fixation points.
• Behavioral measurements of orientation perception.
• Control experiments to rule out alternative explanations (e.g., reporting bias, adaptation).
• Development and application of a Bayesian decoding framework.

Main Results:

• Observed cross-fixation interactions (repulsion and correlation) between orientations.
• Demonstrated that task-relevant orientations interact even when spatially separated.
• Ruled out alternative explanations for the observed interactions.
• Successfully explained data using a high-to-low-level Bayesian decoding model.

Conclusions:

• Visual decoding in working memory may often follow a high-to-low-level hierarchy.
• Interactions between lower-level features are constrained by higher-level representations.
• Findings challenge traditional models of visual processing and perception.
• A retrospective Bayesian decoding framework can explain these complex interactions.