Quantifying the Internal Structure of Categories Using a Neural Typicality Measure

Abstract

How categories are represented continues to be hotly debated across neuroscience and psychology. One topic that is central to cognitive research on category representation but underexplored in neurobiological research concerns the internal structure of categories. Internal structure refers to how the natural variability between-category members is coded so that we are able to determine which members are more typical or better examples of their category. Psychological categorization models offer tools for predicting internal structure and suggest that perceptions of typicality arise from similarities between the representations of category members in a psychological space. Inspired by these models, we develop a neural typicality measure that allows us to measure which category members elicit patterns of activation that are similar to other members of their category and are thus more central in a neural space. Using an artificial categorization task, we test how psychological and physical typicality contribute to neural typicality, and find that neural typicality in occipital and temporal regions is significantly correlated with subjects' perceptions of typicality. The results reveal a convergence between psychological and neural category representations and suggest that our neural typicality measure is a useful tool for connecting psychological and neural measures of internal category structure.

What do differences between multi-voxel and univariate analysis mean? How subject-, voxel-, and trial-level variance impact fMRI analysis

“…Often times it is important to also know the content of these dimensions or what representational or process-level information they contain. Multidimensional scaling (MDS) techniques can be used either in isolation (Davis and Poldrack, 2013b; Diedrichsen et al 2011; Kriegeskorte et al, 2008; Liang, Wagner, & Preston, 2013) or in tandem with Diedrichsen et al’s (2013) analytic dimensionality solution to uncover the content of the dimensions underlying classifier performance or similarity analysis. MDS techniques project the similarities or classification confusion matrices between stimuli/conditions onto a lower dimensional space that can be more easily visualized.…”

Representational geometry: integrating cognition, computation, and the brain

“…Instead, the ventral temporal code might be optimized to emphasize behaviorally relevant categorical divisions and semantic dimensions [44–49]. The geometric centrality of an object in the representation has been linked to the perception of typicality [49]. Animates appear to form a representational cluster not only in IT but also in the amygdala [50].…”

Progress and challenges in probing the human brain

“…Whereas MVPA is generally used to decode individual psychological states, RSA instead asks how the patterns of brain activity evoked by different stimuli are related to one another, and thus provides the means to directly address questions of how mental representations are implemented in the brain. RSA has enabled the demonstration of direct isomorphisms between psychological representations of stimuli (such as the similarity or typicality of objects) and the neural patterns associated with those stimuli 25,26 . Because psychological theories often make predictions regarding the similarity of different stimuli, RSA has also enabled the direct testing of theories, such as theories about how categories are represented 27 and theories of how repeated experiences lead to enhanced learning 28 .…”