Generative Models for Active Vision.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Vision Science

Background:

• The active visual system involves visual cortices, attention networks, and oculomotor control.
• Active inference is a key framework for understanding sensorimotor systems, including vision.
• It posits the brain uses generative models to predict sensory input.

Purpose of the Study:

• To overview generative models employed by the brain for active vision.
• To specify processes explaining retinal and oculomotor proprioceptive activity.
• To explore the relationship between "looking" and "seeing" within these models.

Main Methods:

• Review of active inference principles applied to the visual system.
• Analysis of generative models predicting visual and proprioceptive input.
• Examination of how eye movements and belief updating contribute to perception.

Main Results:

• Generative models are crucial for predicting retinal and oculomotor signals.
• Active vision involves actions (eye movements) to match sensations with predictions.
• Visual perception arises from updating predictions to align with sensory evidence.

Conclusions:

• Active inference provides a unified account of sensorimotor control in vision.
• Understanding generative models is key to deciphering active vision and perception.
• The interplay between action and perception is central to the brain's visual processing.