Bayesian models of eye movement selection with retinotopic maps

Area of Science:

• Neuroscience
• Computational Vision
• Psychophysics

Background:

• Eye movement selection is guided by various criteria.
• Peripheral vision has limited resolution, potentially influencing eye movements.
• Object tracking in daily life requires efficient visual guidance.

Purpose of the Study:

• To investigate the role of position uncertainty in peripheral vision for eye movement selection.
• To test the hypothesis that reducing uncertainty is a key principle in everyday object tracking.
• To compare computational models with human psychophysical data.

Main Methods:

• Developed Bayesian computational models of eye movement selection with increasing complexity.
• Designed a freely moving eye version of the multiple object tracking task.
• Recorded human eye movements during the psychophysical task.

Main Results:

• Computational models incorporating uncertainty showed improved prediction of human eye movements.
• The models' layered structures were inspired by neurobiology.
• Explicitly accounting for uncertainty significantly enhanced predictive performance.

Conclusions:

• Position uncertainty in peripheral vision plays a crucial role in guiding eye movements.
• The principle of uncertainty reduction is likely integral to eye movement selection during object tracking.
• Bayesian models that account for uncertainty offer a more accurate framework for understanding eye movement control.