A Model of the Superior Colliculus Predicts Fixation Locations during Scene Viewing and Visual Search

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Computational models of attention predict fixations using saliency and target maps.
• Existing models lack integration with oculomotor system constraints and real-world task testing.
• Saccade programming models are neurophysiologically detailed but not tested on complex stimuli.

Purpose of the Study:

• To develop a novel computational model of attention that integrates oculomotor constraints with biologically plausible mechanisms.
• To test the model's ability to predict human fixation locations in naturalistic scenes and during search tasks.
• To leverage principles of superior colliculus (SC) organization for improved attention modeling.

Main Methods:

• Developed the MASC (Model of Attention in the Superior Colliculus) model, incorporating neurophysiological constraints of saccade programming.
• Utilized principles of SC organization: foveal over-representation, size-invariant codes, cascaded averaging, and competitive motor maps.
• Evaluated MASC's predictive performance against state-of-the-art models using human eye-tracking data from scene viewing and search tasks.

Main Results:

• MASC accurately predicted human fixation locations across diverse tasks, including free viewing of natural scenes and exemplar/categorical search.
• The model's performance was comparable or superior to specialized, existing state-of-the-art attention models.
• MASC's success is attributed to its incorporation of core SC organizational principles.

Conclusions:

• The MASC model demonstrates the value of incorporating brain-inspired principles, specifically SC organization, into computational models of attention.
• This brain-inspired approach enhances the prediction of overt attention movements in complex, real-world scenarios.
• MASC provides a foundation for generating testable predictions of neural and behavioral responses, advancing autonomous systems research.