Category Representation for Classification and Feature Inference.

Abstract

This research's purpose was to contrast the representations resulting from learning of the same categories by either classifying instances or inferring instance features. Prior inference learning research, particularly T. Yamauchi and A. B. Markman (1998), has suggested that feature inference learning fosters prototype representation, whereas classification learning encourages exemplar representation. Experiment 1 supported this hypothesis. Averaged and individual participant data from transfer after inference training were better fit by a prototype than by an exemplar model. However, Experiment 2, with contrasting inference learning conditions, indicated that the prototype model was mimicking a set of label-based bidirectional rules, as determined by the inference learning task demands in Experiment 1. Only the set of rules model accounted for all the inference learning conditions in these experiments.

Classification versus inference learning contrasted with real-world categories

“…If so, then categorylearning theories based only on classification learning (almost all current theories) may not provide a strong foundation for a general understanding of category learning. Rehder, Colner, and Hoffman (2009) differentiated three explanations for the classification-inference performance differences (which range from inherent task differences to differences due to particulars of the paradigms): the category-centered learning hypothesis (e.g., Markman & Ross, 2003;Yamauchi & Markman, 1998), anticipatory learning (Rehder et al, 2009), and the set-of-rules model (Johansen & Kruschke, 2005). We describe these three explanations below and then propose how they may be tested.…”

“…A. L. Anderson et al (2002) found that inference learners who only inferred two out of four features still learned more about the nonqueried features than did classification learners. However, Rehder et al (2009) showed that when participants are explicitly told they will only be asked about some particular features, their looking time to nonqueried features decreases drastically (although the Rehder et al features were not part of the usual integrated figures, but were spatially separated) Third, the set-of-rules model proposes that, because of the peculiarities of the studies, inference learners simply learn associations between the category label and each feature (Johansen & Kruschke, 2005). According to this account, the inference-classification performance differences can be explained by two differences in how category structures were implemented in previous studies.…”

The costs of supervised classification: The effect of learning task on conceptual flexibility.

“…Some researchers have argued that inference learning motivates people to learn what the categories are like, with attention to within-category feature correlations and prototypical information Markman & Ross, 2003;Yamauchi & Markman, 1998), an account that predicts that learners will fixate most features on most trials, regardless of whether they are ever queried. Others have suggested that feature inference learning simply involves acquiring rule-like associations between the category label and the features (Johansen & Kruschke, 2005; also see Nilsson & Olsson, 2005;Sweller & Hayes, in press), an account that predicts that only the category label and the to-be-predicted dimension are fixated. In contrast, the anticipatory learning account specifies that learners will study that category information they think will be queried about in the future, and this is just what we found in Experiment 2.…”

Category labels versus feature labels: Category labels polarize inferential predictions

“…There was a main effect of label condition [F 1 (3,207) where N i N N represents the number of participants in the ith label condition). We measured the proximity distances of all pairs of individual response vectors in each label condition and applied a hierarchical cluster analysis to these response vectors (see Johansen & Kruschke, 2005, for a similar analysis). The logic behind this analysis was that if individual response patterns in a given label condition were homogeneous, these response vectors should be highly similar to each other, resulting in a large cluster of response vectors.…”