Investigations of exemplar and decision bound models in large, ill-defined category structures.

Experiments involving large-size, ill-defined categories were conducted to distinguish between the predictions of an exemplar model and linear and quadratic decision bound models. In conditions in which the optimal classification boundary was of a more complex form than the quadratic model, the exemplar model provided significantly better accounts of study participants' data than did the decision bound models, even in situations in which a linear bound would have yielded nearly optimal performance. The results suggest that participants are not predisposed or constrained to use linear or quadratic decision bounds for classifying multidimensional perceptual stimuli and that exemplar models may provide a parsimonious process-level account of the complex types of decision bounds used by experiment participants. The results also suggest some limitations on the complexity of the decision bounds that can be learned, in contrast to the predictions of the exemplar model. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Inferring the characteristics of other people: Categories and attribute accessibility.

Investigated how the social inference process operates once a perceiver has activated a particular social category. Three models were tested in an RT paradigm. These models make differing predictions about how the degree of category–attribute association affects the speed and confidence with which an inference is made. In Phase 1, 37 undergraduates assessed the degree of association between a series of life-style attributes and 2 occupational prototypes. In Phase 2, 12 Ss made yes–no judgments about whether these attributes were true of a prototypical category member, and their response latencies and confidence levels were recorded. The "differential accessibility plus contradictions" model was strongly supported. This model states that perceivers make affirming inferences more quickly and confidently about highly associated attributes. When perceivers infer that an attribute is not true of a typical category member, low-associate characteristics can be rejected more quickly and confidently. Inter-S agreement was greater for highly associated attributes. Implications for social perception are discussed. (24 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

A model of probabilistic category learning.

A new connectionist model (named RASHNL) accounts for many "irrational" phenomena found in nonmetric multiple-cue probability learning, wherein people learn to utilize a number of discrete-valued cues that are partially valid indicators of categorical outcomes. Phenomena accounted for include cue competition, effects of cue salience, utilization of configural information, decreased learning when information is introduced after a delay, and effects of base rates. Exps 1 and 2 replicate previous experiments on cue competition and cue salience, and fits of the model provide parameter values for making qualitatively correct predictions for many other situations. The model also makes 2 new predictions, confirmed in Exps 3 and 4. The model formalizes 3 explanatory principles: rapidly shifting attention with learned shifts, decreasing learning rates, and graded similarity in exemplar representation. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Exemplar accounts of blending and distinctiveness effects in perceptual old–new recognition.

Recent findings from the perceptual old–new recognition literature indicate that observers have extremely high false-alarm rates to new items that are "blends" of old ones. In addition, evidence suggests that "distinctive" old items—that is, those located in isolated regions of the similarity space—are recognized with higher probability than are typical old items. Both types of phenomena challenge the predictions of global-familiarity exemplar models of perceptual old–new recognition, which posit that the probability that an observer judges an item as old is based on its summed similarity to previously presented exemplars. In the present research the authors pursued these blending and distinctiveness effects by testing paradigms in which similarity relations among objects are highly controlled and in which the variables of blending and distinctiveness are not confounded with other properties associated with the individual objects themselves. In contrast to previous results, the authors found effects of blending and distinctiveness that are compatible with the predictions of a pure summed-similarity exemplar model. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Prototypes in category learning: The effects of category size, category structure, and stimulus complexity.

Although research in categorization has sometimes been motivated by prototype theory, recent studies have favored exemplar theory. However, some of these studies focused on small, poorly differentiated categories composed of simple, 4-dimensional stimuli. Some analyzed the aggregate data of entire groups. Some compared powerful multiplicative exemplar models to less powerful additive prototype models. Here, comparable prototype and exemplar models were fit to individual-participant data in 4 experiments that sampled category sets varying in size, level of category structure, and stimulus complexity (dimensionality). The prototype model always fit the observed data better than the exemplar model did. Prototype-based processes seemed especially relevant when participants learned categories that were larger or contained more complex stimuli. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Exemplar effects in categorization and multiple-cue judgment.

Categorization and multiple-cue judgment are similar tasks, but the influential models in the two areas are different in terms of the computations, processes, and neural substrates that they imply. In categorization, exemplar memory is often emphasized, whereas multiple-cue judgment generally is interpreted in terms of integration of Cues that have been abstracted in training. In 3 experiments the authors investigated whether these conclusions derive from genuine differences in the processes or are accidental to the different research methods. The results revealed large individual differences and a shift from exemplar memory to cue abstraction when the criterion is changed from a binary to a continuous variable, especially for a probabilistic criterion. People appear to switch between qualitatively distinct processes in the 2 tasks. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Perceptual learning of spatial localization: specificity for orientation, position, and context

Discrimination of simple visual attributes can improve significantly with practice. We have trained human observers to perform peripherally presented tasks involving the localization of short line segments and examined the specificity of the learning for the visual location, orientation, and geometric arrangement of the trained stimulus. Several weeks of training resulted in dramatic threshold reductions. The learning was specific for the orientation and location of the trained stimulus, indicating the involvement of the earliest cortical stages in the visual pathway where the orientation and location of stimuli are mapped with highest resolution. Furthermore, improvement was also specific for both the configuration of the trained stimulus and the attribute of the stimulus that was under scrutiny during training. This degree of specificity suggests that the learning cannot be achieved by cortical recruitment alone, as proposed in current models, but is likely to involve a refinement of lateral interactions within the cortex and possibly a gating of lower level changes by attentional mechanisms.     

Models of quantitative estimations: Rule-based and exemplar-based processes compared.

The cognitive processes underlying quantitative estimations vary. Past research has identified task-contingent changes between rule-based and exemplar-based processes (P. Juslin, L. Karlsson, & H. Olsson, 2008). B. von Helversen and J. Rieskamp (2008), however, proposed a simple rule-based model—the mapping model—that outperformed the exemplar model in a task thought to promote exemplar-based processing. This raised questions about the assumptions of rule-based versus exemplar-based models that underlie the notion of task contingency of cognitive processes. Rule-based models, such as the mapping model, assume the abstraction of explicit task knowledge. In contrast, exemplar models should profit if storage and activation of the exemplars is facilitated. Two studies tested the importance of the two models’ assumptions. When knowledge about cues existed, the rule-based mapping model predicted quantitative estimations best. In contrast, when knowledge about the cues was difficult to gain, participants’ estimations were best described by an exemplar model. The results emphasize the task contingency of cognitive processes. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Prototype and Exemplar Accounts of Category Learning and Attentional Allocation: A Reassessment.

In a recent article. J. P. Minda and J. D. Smith (2002; see record 2002-00620-002) argued that an exemplar model provided worse quantitative fits than an alternative prototype model to individual subject data from the classic D. L. Medin and M. M. Schaffer (1978) 5/4 categorization paradigm. In addition, they argued that the exemplar model achieved its fits by making untenable assumptions regarding how observers distribute their attention. In this article, we demonstrate that when the models are equated in terms of their response-rule flexibility, the exemplar model provides a substantially better account of the categorization data than does a prototype or mixed model. In addition, we point to shortcomings in the attention-allocation analyses conducted by J. P. Minda and J. D. Smith (2002). When these shortcomings are corrected, we find no evidence that challenges the attention-allocation assumptions of the exemplar model. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Between ignorance and truth: Partition dependence and learning in judgment under uncertainty.

In 3 studies, participants viewed sequences of multiattribute objects (e.g., colored shapes) appearing with varying frequencies and judged the likelihood of the attributes of those objects. Judged probabilities reflected a compromise between (a) the frequency with which each attribute appeared and (b) the ignorance prior probability cued by the number of distinct values that the focal attribute could take on. Thus, judged probabilities were partition dependent, varying with the number of events into which the state space was subjectively divided. This bias was diminished among participants more confident in what they learned, was strong and insensitive to level of confidence when ignorance priors were especially salient, and required ignorance priors to be salient only when probabilities were elicited (not during encoding). (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Cue discovery in probabilistic environments: Uncertainty and experimentation.

It is argued that cue discovery is an important part of the task of learning from experience in probabilistic environments. Cue discovery involves the identification of new valid predictive cues, and their incorporation into the learner's mental model. In contrast, previous work in "multiple-cue probability learning" has focused principally on learning the quantitative characteristics of given cue–criterion relations. In two studies presented here, 36 college students learned to predict the behavior of a computer-controlled graphic display. The criterion behavior was a function of four to six cues, including several not explicitly identified. Analyses of test results, verbal reports, and behavior during learning indicate that subjects used outcome feedback to accomplish cue discovery. This process was accomplished more effectively when subjects were permitted to experiment by designing their own learning experiences. A distinction is drawn between a system whose behavior is unpredictable because of the presence of unknown controlling variables, and one that contains "random" error. Cue discovery may be more difficult in the latter case, but is accomplished nonetheless. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Rules and exemplars in category learning

Psychological theories of categorization generally focus on either rule- or exemplar-based explanations. We present 2 experiments that show evidence of both rule induction and exemplar encoding as well as a connectionist model, ATRIUM, that specifies a mechanism for combining rule- and exemplar-based representation. In 2 experiments participants learned to classify items, most of which followed a simple rule, although there were a few frequently occurring exceptions. Experiment 1 examined how people extrapolate beyond the range of training. Experiment 2 examined the effect of instance frequency on generalization. Categorization behavior was well described by the model, in which exemplar representation is used for both rule and exception processing. A key element in correctly modeling these results was capturing the interaction between the rule- and exemplar-based representations by using shifts of attention between rules and exemplars.     

Base rates can affect individual predictions.

356 college students in 4 experiments learned base-rate information and then participated in a predictive social judgment task. In all experiments, predictive behavior was significantly affected by the relevant base rates. In Exp II, for example, Ss in a high base-rate condition were more likely to predict that a given student planned to enroll for further training than were Ss in a low base-rate condition. The base-rate variable had a greater influence than might have been anticipated from the Ss' confidence ratings or from their replies to a series of postexperimental questions concerning the cues they had relied on in formulating their predictions. Nonetheless, the base-rate data were substantially less influential than a Bayesian model would require. In contrast to D. Kahneman and A. Tversky's (1973) analysis, individuating information that reflected a high (vs low) degree of representativeness did not determine the magnitude of the base-rate effect. (22 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Attention and luminance detection: A quantitative analysis.

Two experiments investigated the effects of spatial cues on the detection of luminance targets in global (nonlocalizing) and local (poststimulus probe) detection tasks. Cuing effects were quantified by fitting a variety of single- and multiple-channel signal-detection models to the data from individual observers. The local detection data showed enhanced sensitivity at cued locations, consistent with previous findings. Models for global detection in which cues affected both sensitivity and criterion or criterion alone could not be distinguished reliably on the basis of goodness-of-fit statistics. However, estimated model parameters showed that under global detection conditions, sensitivity was enhanced at cued locations by a factor of around 2. Results are discussed relative to a model in which Poisson-coded sensory representations are modulated by spatial attention. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Contrasting predictions of extended comparator hypothesis and acquisition-focused models of learning concerning retrospective revaluation.

Three conditioned suppression experiments with rats investigated contrasting predictions made by the extended comparator hypothesis and acquisition-focused models of learning, specifically, modified SOP and the revised Rescorla–Wagner model, concerning retrospective revaluation. Two target cues (X and Y) were partially reinforced using a stimulus relative validity design (i.e., AX–Outcome; BX–No outcome; CY–Outcome; DY–No outcome), and subsequently one of the companion cues for each target was extinguished in compound (BC–No outcome). In Experiment 1, which used spaced trials for relative validity training, greater suppression was observed to target cue Y for which the excitatory companion cue had been extinguished in relation to target cue X for which the nonexcitatory companion cue had been extinguished. Experiment 2 replicated these results in a sensory preconditioning preparation. Experiment 3 massed the trials during relative validity training, and the opposite pattern of data was observed. The results are consistent with the predictions of the extended comparator hypothesis. Furthermore, this set of experiments is unique in being able to differentiate between these models without invoking higher-order comparator processes. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The limitations of decision trees and automatic learning in real world medical decision making

The decision tree approach is one of the most common approaches in automatic learning and decision making. The automatic learning of decision trees and their use usually show very good results in various "theoretical" environments. But in real life it is often impossible to find the desired number of representative training objects for various reasons. The lack of possibilities to measure attribute values, high cost and complexity of such measurements, and unavailability of all attributes at the same time are the typical representatives. For this reason we decided to use the decision trees not for their primary task--the decision making--but for outlining the most important attributes. This was possible by using a well-known property of the decision trees--their knowledge representation, which can be easily understood by humans. In a delicate field of medical decision making, we cannot allow ourselves to make any inaccurate decisions and the "tips," provided by the decision trees, can be of a great assistance. Our main interest was to discover a predisposition to two forms of acidosis: the metabolic acidosis and respiratory acidosis, which can both have serious effects on child's health. We decided to construct different decision trees from a set of training objects. Instead of using a test set for evaluation of a decision tree, we asked medical experts to take a closer look at the generated trees. They examined and evaluated the decision trees branch by branch. Their comments show that trees generated from the available training set mainly have surprisingly good branches, but on the other hand, for some, no medical explanation could be found.     

Song-note discriminations in zebra finches (Taeniopygia guttata): Categories and pseudocategories.

Sturdy, Phillmore, and Weisman (1999) developed a simple, reliable procedure for classifying zebra finch (Taeniopygia guttata) song notes into types using exemplar notes from 4 captive North American laboratory colonies. In the present experiment, zebra finches learned true-category song-note discriminations based on Sturdy et al.'s note-type classification faster than pseudocategory discriminations based on random assortments of the same notes. Also, birds in the true-category discrimination were able to sort notes not heard during training into the same categories as the training notes. This evidence suggests that zebra finches accurately sort conspecific song notes into the same open-ended categories as Sturdy et al.'s human observers. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The effect of category variability in perceptual categorization.

Exemplar and distributional accounts of categorization make differing predictions for the classification of a critical exemplar precisely halfway between the nearest exemplars of 2 categories differing in variability. Under standard conditions of sequential presentation, the critical exemplar was classified into the most similar, least variable category, consistent with an exemplar account. However, if the difference in variability is made more salient, then the same exemplar is classified into the more variable, most likely category, consistent with a distributional account. This suggests that participants may be strategic in their use of either strategy. However, when the relative variability of 2 categories was manipulated, participants showed changes in the classification of intermediate exemplars that neither approach could account for. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Overconfidence effects in category learning: A comparison of connectionist and exemplar memory models.

Exemplar and connectionist models were compared on their ability to predict overconfidence effects in category learning data. In the standard task, participants learned to classify hypothetical patients with particular symptom patterns into disease categories and reported confidence judgments in the form of probabilities. The connectionist model asserts that classifications and confidence are based on the strength of learned associations between symptoms and diseases. The exemplar retrieval model (ERM) proposes that people learn by storing examples and that their judgments are often based on the first example they happen to retrieve. Experiments 1 and 2 established that overconfidence increases when the classification step of the process is bypassed. Experiments 2 and 3 showed that a direct instruction to retrieve many exemplars reduces overconfidence. Only the ERM predicted the major qualitative phenomena exhibited in these experiments. (PsycINFO Database Record (c) 2010 APA, all rights reserved)