Exploring the Structure of Spatial Representations.

Area of Science:

• Cognitive Science
• Neuroscience
• Computational Psychology

Background:

• Human spatial memory is often conceptualized as cognitive maps.
• Current understanding suggests these maps might be fragmented into local reference frames rather than unitary global representations.
• The underlying principles governing the structure of these cognitive maps remain largely unclear.

Purpose of the Study:

• To propose and validate a computational model for understanding the structure of human spatial memory representations.
• To investigate the role of psychological space clustering in forming cognitive map structures.
• To develop methods for learning dissimilarity metrics that characterize psychological spaces.

Main Methods:

• Formulated methods for learning dissimilarity functions (metrics) from psychological spaces.
• Employed a probabilistic clustering model based on the Bayesian cognition paradigm.
• Analyzed the influence of spatial distance, visual similarity, and functional similarity on spatial memory structure.

Main Results:

• Learned metrics and the clustering model successfully predicted participants' cognitive map structures.
• Demonstrated strong correlations between spatial distance, visual similarity, functional similarity, and grouping probability in spatial representations.
• Provided evidence supporting a clustering-based mechanism in the formation of spatial memory.

Conclusions:

• The structure of navigation space representations arises from clustering within psychological spaces.
• This clustering process, informed by representational geometry and similarity, explains cognitive map formation.
• The developed methods offer insights into human spatial representation learning and potential computational tools.