Common Bayesian models for common cognitive issues.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Neuroscience
• Probability Theory

Background:

• Living and artificial cognitive systems must efficiently perceive, infer, decide, and act using incomplete and uncertain environmental models.
• Traditional symbolic logic is insufficient for addressing these challenges due to its inability to handle uncertainty.
• The subjectivist approach to probability offers a framework to extend logic for dealing with environmental uncertainty.

Purpose of the Study:

• To review common cognitive issues within a unified Bayesian framework.
• To formalize concepts such as ambiguities, fusion, multimodality, conflicts, modularity, hierarchies, and loops using Bayesian models.
• To discuss the assumptions, commonalities, and differences among these Bayesian models.

Main Methods:

• Review of cognitive issues from neuroscience, psychophysics, and robotics literature.
• Formalization of cognitive concepts using a template Bayesian model.
• Analysis and discussion of the properties and variations of the proposed Bayesian models.

Main Results:

• Demonstration that a common Bayesian formalism can encompass diverse cognitive issues like ambiguity, fusion, and multimodality.
• Introduction of template Bayesian models for each cognitive concept.
• Discussion of the underlying assumptions and distinctions between different Bayesian approaches to cognitive modeling.

Conclusions:

• The subjectivist approach to probability, formalized within a Bayesian framework, provides a powerful tool for cognitive systems to operate efficiently under uncertainty.
• Bayesian modeling offers a unified approach to understanding and addressing complex cognitive challenges.
• This work highlights the utility of Bayesian methods in bridging perception, inference, decision-making, and action in cognitive systems.