Humans predict the forest, not the trees: statistical learning of spatiotemporal structure in visual scenes

Area of Science:

• Cognitive Neuroscience
• Visual Perception
• Human Brain Function

Background:

• The human brain uses statistical regularities to predict future sensory information.
• Real-world scenes contain multiple objects, posing a challenge for prediction.
• It is unclear if prediction relies on individual object features or scene-level information.

Purpose of the Study:

• To investigate whether perceptual anticipation relies on lower-level (object-specific) or higher-level (scene-specific) information.
• To determine if the brain anticipates individual objects or the scene as a whole.

Main Methods:

• Participants learned associations between co-occurring objects in fixed spatial arrangements.
• Implicit learning of temporal regularities between object displays was assessed.
• Behavioral responses and neural activity (fMRI) were measured following spatial and temporal structure violations.

Main Results:

• A behavioral advantage for temporal regularities was observed only when spatial structures were preserved.
• Neural responses in the lateral occipital cortex showed suppression for expected stimuli within expected configurations.
• These findings indicate that temporal expectations are configuration-specific.

Conclusions:

• The human brain forms expectations about object configurations, prioritizing higher-level information.
• Perceptual anticipation is driven by scene-level predictions rather than individual object predictions.
• This highlights the brain's capacity for complex, holistic predictive processing.