Computational approaches to the development of perceptual expertise

Dog experts, ornithologists, radiologists and other specialists are noted for their remarkable abilities at categorizing, identifying and recognizing objects within their domain of expertise. A complete understanding of the development of perceptual expertise requires a combination of thorough empirical research and carefully articulated computational theories that formalize specific hypotheses about the acquisition of expertise. A comprehensive computational theory of the development of perceptual expertise remains elusive, but we can look to existing computational models from the object-recognition, perceptual-categorization, automaticity and related literatures for possible starting points. Arguably, hypotheses about the development of perceptual expertise should first be explored within the context of existing computational models of visual object understanding before considering the creation of highly modularized adaptations for particular domains of perceptual expertise.     

Introduction to the special section on theory and data in categorization: Integrating computational, behavioral, and cognitive neuroscience approaches

This special section brings together behavioral, computational, mathematical, and neuroimaging approaches to understand the processes underlying category learning. Over the past decade, there has been growing convergence in research on categorization, with computational-mathematical models influencing the interpretation of brain imaging and neuropsychological data, and with cognitive neuroscience findings influencing the development and refinement of models. Classic debates between single-system and multiple-memory-system theories have become more nuanced and focused. Multiple brain areas and cognitive processes contribute to categorization, but theories differ markedly in whether and when those neurocognitive components are recruited for different aspects of categorization. The articles in this special section approach this issue from several diverse angles.     

Bayesian inference with Stan: A tutorial on adding custom distributions

When evaluating cognitive models based on fits to observed data (or, really, any model that has free parameters), parameter estimation is critically important. Traditional techniques like hill climbing by minimizing or maximizing a fit statistic often result in point estimates. Bayesian approaches instead estimate parameters as posterior probability distributions, and thus naturally account for the uncertainty associated with parameter estimation; Bayesian approaches also offer powerful and principled methods for model comparison. Although software applications such as WinBUGS (Lunn, Thomas, Best, & Spiegelhalter, Statistics and Computing, 10, 325–337, 2000) and JAGS (Plummer, 2003) provide “turnkey”-style packages for Bayesian inference, they can be inefficient when dealing with models whose parameters are correlated, which is often the case for cognitive models, and they can impose significant technical barriers to adding custom distributions, which is often necessary when implementing cognitive models within a Bayesian framework. A recently developed software package called Stan (Stan Development Team, 2015) can solve both problems, as well as provide a turnkey solution to Bayesian inference. We present a tutorial on how to use Stan and how to add custom distributions to it, with an example using the linear ballistic accumulator model (Brown & Heathcote, Cognitive Psychology, 57, 153–178. doi: 10.1016/j.cogpsych.2007.12.002, 2008).     

Learning to classify integral-dimension stimuli

The authors tested 288 participants in the classic category-learning tasks introduced by Shepard, Hovland, and Jenkins (1961). However, separable-dimension stimuli were used in previous tests, whereas integral-dimension stimuli were used in the present study. In contrast to previous results, which showed a superiority for Problem Type II over Problem Types III, IV, and V, the reverse pattern was observed in the present research. This result confirms a fundamental prediction made by modern exemplar-based models of classification learning. The results are interpreted in terms of the extent to which selective-attention learning mechanisms operate when separable-dimension versus integral-dimension stimuli are used.     

Comparing exemplar-retrieval and decision-bound models of speeded perceptual classification

The authors compared the exemplar-based random-walk (EBRW) model of Nosofsky and Palmeri (1997) and the decision-bound model (DBM) of Ashby and Maddox (1994; Maddox & Ashby, 1996) on their ability to predict performance in Garner’s (1974) speeded classification tasks. A key question was the extent to which the models could predict facilitation in the correlated task and interference in the filtering task, in situations involving integral-dimension stimuli. To obtain rigorous constraints for model evaluation, the goal was to fit the detailed structure of the response time (RT) distribution data associated with each individual stimulus in each task. Both models yielded reasonably good global quantitative fits to the RT distribution and accuracy data. However, the DBM failed to properly characterize the interference effects in the filtering task. Apparently, a fundamental limitation of the DBM is that it predicts that the fastest RTs in the filtering task should be faster than the fastest RTs in the control task, whereas the opposite pattern was observed in our data.     

The role of background knowledge in speeded perceptual categorization

We examined the time-course of the influence of background knowledge on perceptual categorization by manipulating the meaningfulness of labels associated with categories and by manipulating the amount of time provided to subjects for making a categorization decision. Extending a paradigm originally reported by Wisniewski and Medin (1994) (Cog. Sci. 18 (1994) 221), subjects learned two categories of children's drawings that were given either meaningless labels (drawings by children from 'group 1' or 'group 2') or meaningful labels (drawings by 'creative' or 'non-creative' children); the meaningfulness of the label had a significant effect on how new drawings were classified. In addition, half of the subjects were provided unlimited time to respond, while the other half of the subjects were forced to respond quickly; speeded response conditions had a relatively large effect on categorization decisions by subjects given the meaningless labels but had relatively little effect on categorization decisions by subjects given the meaningful labels. These results suggest that some forms of background knowledge can show an influence at relatively early stages in the time-course of a categorization decision.     

Learning About Categories in the Absence of Training: Profound Amnesia and the Relationship Between PerceptualCategorization and Recognition Memory

Previous evidence suggests that amnesics can categorize stimuli as well as normal individuals but are significantly worse at recognizing those stimuli. In an extreme case, a profoundly amnesic individual, E.P., was found to have near-normal categorization, yet, unlike most amnesics, was unable to recognize better than chance. This evidence has been used to argue against the possibility that a common memory system underlies these cognitive processes. However, we provide evidence that the experimental procedures typically used to test amnesic individuals may be flawed in that initial exposure to category members may be unnecessary to observe accurate categorization of test stimuli. We experimentally "induced" profound amnesia in normal individuals by telling them they had viewed subliminally presented stimuli, which were never actually presented. Using the same experimental paradigm used to test amnesics, we observed that participants' recognition performance was completely at chance, as should be expected, yet categorization performance was quite good.     

Learning categories at different hierarchical levels: a comparison of category learning models

Three formal models of category learning, the rational model (Anderson, 1990), the configural-cue model (Gluck & Bower, 1988a), and ALCOVE (Kruschke, 1992), were evaluated on their ability to account for differential learning of hierarchically structured categories. An experiment using a theoretically challenging category structure developed by Lassaline, Wisniewski, and Medin (1992) is reported. Subjects learned one of two different category structures. For one structure, diagnostic information was present along a single dimension (1-D). For the other structure, diagnostic information was distributed across four dimensions (4-D). Subjects learned these categories at a general or at a specific level of abstraction. For the 1-D structure, specific-level categories were learned more rapidly than general-level categories. For the 4-D structure, the opposite result was observed. These results proved highly diagnostic for evaluating the models—although ALCOVE provided a good account of the observed results, the rational model and the configural-cue model did not.