Invited Talk: Evaluation of a computational framework for cognitive maps

For further information please click here.

Abstract:

The presence or absence of cognitive maps in spatial representation has been intensely debated in the literature for the past 50 years. However, there is still no clear consensus regarding under what conditions a navigator may acquire a cognitive map. A large part of this challenge is attributable to variability in the ways in which cognitive maps have been operationalized. Cognitive maps may be defined with respect to an allocentric reference frame (e.g., Lloyd, 1989), large scale (e.g., Shettleworth, 2010), Euclidean metric (e.g., Golledge & Hubert, 1982), or high level of abstraction (e.g., Portugali, 1996). While some researchers have argued that the term “cognitive map” should be eliminated (e.g., Tversky, 1993), formalizing the term may help disentangle these various aspects. At this workshop in 2016, I presented a first attempt at such formalization using a variation of the Generalized Context Model (Nosofsky, 1986). During the 2017 workshop, I will present modifications of this cognitive map framework and evidence for and against our approach to modeling different reference frames. This validation step involves data from multiple published and unpublished experiments. All of these datasets have been used to classify different sets of spatial responses (each towards a specific arrangement of locations) as reflecting either egocentric or allocentric reference frames. Supervised machine learning was then used to derive the reference points from which the observers represented the original arrangements of locations. Future work will be focused on validating other aspects of this modeling framework (i.e., scale, metric, and level of abstraction) and then devising predictions that would otherwise remain largely untestable.