Scene Construction, Visual Foraging, and Active Inference.

Area of Science:

• Computational Neuroscience
• Cognitive Science
• Neuroscience

Background:

• Active inference posits that perception and action minimize variational free energy.
• This framework integrates risk-sensitive control and expected utility theory with epistemic value.
• Epistemic value quantifies the importance of information in resolving uncertainty.

Purpose of the Study:

• To describe an active inference scheme for visual search and scene construction.
• To apply active inference to saccadic visual scene categorization.
• To link approximate Bayesian inference with functional segregation in the visual cortex.

Main Methods:

• Modeling active inference for saccadic visual search.
• Considering scene categorization based on spatial object relationships.
• Utilizing a sequential evidence accumulation process with prediction and postdiction.

Main Results:

• The proposed scheme highlights the functional anatomy connecting Bayesian inference and visual cortex organization.
• Demonstrates how active inference can model the sequential sampling of visual cues.
• Provides a neurobiologically plausible process theory for active inference in Markov decision processes.

Conclusions:

• Active inference offers a unified framework for understanding perception and action in visual search.
• The model elucidates the relationship between approximate Bayesian inference and visual cortex functional segregation.
• Future work will model empirical saccadic searches and explore intersubject variability in information foraging, including in conditions like schizophrenia.