Abstract
Human categorization is one of the most important and successful targets of cognitive modeling, with decades of model development and assessment using simple, low-dimensional artificial stimuli. However, it remains unclear how these findings relate to categorization in more natural settings, involving complex, high-dimensional stimuli. Here, we take a step towards addressing this question by modeling human categorization over a large behavioral dataset, comprising more than 500,000 judgments over 10,000 natural images from ten object categories. We apply a range of machine learning methods to generate candidate representations for these images, and show that combining rich image representations with flexible cognitive models captures human decisions best. We also find that in the high-dimensional representational spaces these methods generate, simple prototype models can perform comparably to the more complex memory-based exemplar models dominant in laboratory settings.
Highlights
Human categorization is one of the most important and successful targets of cognitive modeling, with decades of model development and assessment using simple, lowdimensional artificial stimuli
This work has been insightful and theoretically productive, we know little about how it relates to the complex visual world it was meant to describe: the focus on designing experiments to distinguish between models means that it derives almost exclusively from studies using highly controlled and simplified perceptual stimuli, represented mathematically by lowdimensional hand-coded descriptions of obvious features or low-dimensional multidimensional-scaling (MDS) solutions based on similarity judgments[11,12,17,22,23,24,25,26,27,28,29,30,31]
We find that choice of feature representation affects the predictive performance of categorization models profoundly, which cognitive models are of most benefit over their machine learning counterparts for ambiguous images, and that there is little difference in the performance of prototype and exemplar strategies in the types of high-dimensional representational spaces that support natural image categorization best
Summary
Human categorization is one of the most important and successful targets of cognitive modeling, with decades of model development and assessment using simple, lowdimensional artificial stimuli. This work has been insightful and theoretically productive, we know little about how it relates to the complex visual world it was meant to describe: the focus on designing experiments to distinguish between models means that it derives almost exclusively from studies using highly controlled and simplified perceptual stimuli, represented mathematically by lowdimensional hand-coded descriptions of obvious features or low-dimensional multidimensional-scaling (MDS) solutions based on similarity judgments[11,12,17,22,23,24,25,26,27,28,29,30,31] (see Fig. 1) Extending these findings to more realistic settings, and in particular to natural images, remains a central challenge. The effect of increasing the dimensionality of stimulus representations on categorization model performance is largely unexplored and yet likely to be a key factor in adequately representing more complex natural stimuli
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
