Errors, efficiency, and the interplay between attention and category learning

Learning to identify objects as members of categories is an essential cognitive skill and learning to deploy attention effectively is a core component of that process. The present study investigated an assumption imbedded in formal models of categorization: error is necessary for attentional learning. Eye-trackers were used to record participants’ allocation of attention to task relevant and irrelevant features while learning a complex categorization task. It was found that participants optimized their fixation patterns in the absence of both performance errors and corrective external feedback. Optimization began immediately after each category was mastered and continued for many trials. These results demonstrate that error is neither necessary nor sufficient for all forms of attentional learning.     

MODELING PROBABIUSTIC CATEGORIZATION DATA:

Abstract— In probabilistic categorization tasks, the correct category is determined only probabilistically by the stimulus pattern Data from such experiments have been successfully accounted for by a simple network model, but have posed difficulties for exemplar models In the present article, we consider an exemplar model, CLEM (concept learning by exemplar memorization), which differs from previously tested exemplar models in that exemplar traces are assumed to be stored only when the subject has guessed or made a classification error Fits of CLEM to both learning and test data were comparable to those of the network model, and better than those obtained for a version of CLEM in which encoding was independent of the subject's response The implications of these results for the processes underlying classification decisions are discussed.     

Amount of reward has opposite effects on the discounting of delayed and probabilistic outcomes

Previous research has shown that the value of large future rewards is discounted less steeply than is the value of small future rewards. These experiments extended this line of research to probabilistic rewards. Two experiments replicated the standard findings for delayed rewards but demonstrated that amount has an opposite effect on the discounting of probabilistic rewards. That is, large probabilistic amounts were discounted at the same or higher rates than small amounts. Although amount had opposite effects on the discounting of delayed and probabilistic rewards, nevertheless, the same form of mathematical function accurately described discounting of both types of reward. The findings suggest that fundamentally similar, but not identical, processes are involved in decision making regarding delayed and probabilistic rewards. The implications of these findings for impulsivity and self-control are discussed.     

Effects of inflation on the subjective value of delayed and probabilistic rewards

In the years prior to 1994, there were very high rates of inflation in Poland, and the zloty depreciated relative to the U.S. dollar. However, the new zloty, introduced in 1995, was associated with greatly decreased rates of inflation and provided a more stable currency. We report a series of three experiments that take advantage of these changes to examine the effects of inflation on the subjective value of delayed and probabilistic rewards. Subjects were Polish citizens familiar with both zlotys and dollars. The first two experiments, conducted in 1994, used dollars and old zlotys, and the third experiment, conducted in 1996, used dollars and new zlotys. In all three experiments, the dollar and zloty rewards were of equivalent worth, according to the then current exchange rates. In Experiment 1, subjects chose between immediate and delayed rewards and, in Experiment 2, chose between certain and probabilistic rewards. The subjective value of a delayed reward was greater when its amount was specified in dollars than when it was specified in old zlotys. In contrast, the currency in which a reward was specified had no effect on the subjective value of probabilistic rewards. The results of these two experiments suggest a selective effect of inflation on decisions involving delayed rewards. This was verified in the third experiment, in which, using new zlotys, no differences in discounting were observed between the two currencies with either probabilistic or delayed rewards. Importantly, in all three experiments, the discounting of both delayed and probabilistic rewards was well described by the same simple mathematical model, suggesting that similar decision-making processes underlie both phenomena. However, the present results argue against a single-process theory in which the discounting of probabilistic rewards is derived from the discounting of delayed rewards.     

Worse than feared? Failure induction modulates the electrophysiological signature of error monitoring during subsequent learning

This study examined how self-relevant failure influences error monitoring—as reflected in the error-related negativity (Ne/ERN) —and behavioral adaptation during subsequent feedback-based learning. We applied two phases (pre- and posttest) of a probabilistic learning task. Between pre- and posttest, participants were assigned to one of two groups receiving either failure feedback or no feedback during a visual search task described as diagnostic of intellectual abilities. To disentangle the effects of failure and motivational disengagement due to prolonged task performance, we linked the posttest to intelligence (Experiment 1) or described it in neutral terms (Experiment 2). Failure induction was associated with an increase in Ne/ERN amplitude at posttest in both experiments, although there were no differences in overall performance. In contrast, the Ne/ERN decreased from pre- to posttest in the no-failure-feedback group, particularly in Experiment 2. Furthermore, failure feedback affected error-related behavioral adjustments, suggesting a shift toward a reactive, error-driven mode of behavior control. These findings emphasize the importance of affective-motivational state in error processing and subsequent behavioral adaptation.     

Ad hoc category restructuring

Participants learned to classify seemingly arbitrary words into categories that also corresponded to ad hoc categories (see, e.g., Barsalou, 1983). By adapting experimental mechanisms previously used to study knowledge restructuring in perceptual categorization, we provide a novel account of how experimental and preexperimental knowledge interact. Participants were told of the existence of the ad hoc categories either at the beginning or the end of training. When the ad hoc labels were revealed at the end of training, participants switched from categorization based on experimental learning to categorization based on preexperimental knowledge in some, but not all, circumstances. Important mediators of the extent of that switch were the amount of performance error experienced during prior learning and whether or not prior knowledge was in conflict with experimental learning. We present a computational model of the trade-off between preexperimental knowledge and experimental learning that accounts for the main results.     

Strategy development and learning differences in supervised and unsupervised categorization

The processes that determine unsupervised categorization, the task of classifying stimuli without guidance or feedback, are poorly understood. Two experiments examined the emergence and plasticity of unsupervised strategies using perceptual stimuli that varied along two separable dimensions. In the first experiment, participants either classified stimuli into any two categories of their choice or learned identical classifications by supervised categorization. Irrespective of the complexity of classification, supervised and unsupervised learning rates differed little when both modes of learning were maximally comparable. The second experiment examined the plasticity of unsupervised classifications by introducing novel stimuli halfway through training. Whether or not people altered their strategies, they responded to novel stimuli in a gradual manner. The gradual and continuous evolution and adaptation of strategies suggests that unsupervised categorization involves true learning which shares many properties of supervised category learning. We also show that the choice of unsupervised strategy cannot be predicted from the properties of early learning trials, but is best understood as a function of the initial distribution of dimensional attention.     

Combining advice: the weight of a dissenting opinion in the consensus

We present two studies that evaluate how people combine advice and how they respond to outlying opinions. In a preliminary study, we found that individuals use discounting strategies when they encounter an extreme opinion in a small sample of opinions taken only once (a one‐shot advice‐taking situation). The main study examines the influence of outlying opinions (which may or may not be accurate) within a learning paradigm with feedback. This study shows that it is easy to reinforce a discounting strategy (with feedback) whereas it is more difficult to counteract this default strategy. In the discussion we consider cognitive, statistical, and strategic justifications for discounting opinions, from both theoretical and practical points of view. Copyright © 2004 John Wiley & Sons, Ltd.     

Pigeons' discounting of probabilistic and delayed reinforcers

Pigeons' discounting of probabilistic and delayed food reinforcers was studied using adjusting-amount procedures. In the probability discounting conditions, pigeons chose between an adjusting number of food pellets contingent on a single key peck and a larger, fixed number of pellets contingent on completion of a variable-ratio schedule. In the delay discounting conditions, pigeons chose between an adjusting number of pellets delivered immediately and a larger, fixed number of pellets delivered after a delay. Probability discounting (i.e., subjective value as a function of the odds against reinforcement) was as well described by a hyperboloid function as delay discounting was (i.e., subjective value as a function of the time until reinforcement). As in humans, the exponents of the hyperboloid function when it was fitted to the probability discounting data were lower than the exponents of the hyperboloid function when it was fitted to the delay discounting data. The subjective values of probabilistic reinforcers were strongly correlated with predictions based on simply substituting the average delay to their receipt in each probabilistic reinforcement condition into the hyperboloid discounting function. However, the subjective values were systematically underestimated using this approach. Using the discounting function proposed by Mazur (1989), which takes into account the variability in the delay to the probabilistic reinforcers, the accuracy with which their subjective values could be predicted was increased. Taken together, the present findings are consistent with Rachlin's (Rachlin, 1990; Rachlin, Logue, Gibbon, & Frankel, 1986) hypothesis that choice involving repeated gambles may be interpreted in terms of the delays to the probabilistic reinforcers.     

A Probabilistic Latent Class Model for Assessing Inter-Judge Reliability

Increasingly behavioral researchers are soliciting cognitive responses in addition to standard attitudinal measures when attempting to assess the effects of persuasive communications. The coding of the elicited cognitive responses generally involves some sort of categorization, typically undertaken by independent judges, and the quality of the data is, to a large degree, evaluated in terms of some reliability coefficient which reflects the extent to which the independent judges agreed. The purpose of this paper is to present and illustrate a probabilistic model for assessing inter-judge reliability. The proposed probabilistic model allows one to (a) use formal test statistics to evaluate the extent and character of inter-judge reliability, (b) estimate the assignment error rates and their standard errors, and (c) test for simultaneous agreement for more than two judges. The probabilistic model is operationalized in terms of restricted latent class models.     

Beyond hypercorrection: remembering corrective feedback for low-confidence errors*

Correcting errors based on corrective feedback is essential to successful learning. Previous studies have found that corrections to high-confidence errors are better remembered than low-confidence errors (the hypercorrection effect). The aim of this study was to investigate whether corrections to low-confidence errors can also be successfully retained in some cases. Participants completed an initial multiple-choice test consisting of control, trick and easy general-knowledge questions, rated their confidence after answering each question, and then received immediate corrective feedback. After a short delay, they were given a cued-recall test consisting of the same questions. In two experiments, we found high-confidence errors to control questions were better corrected on the second test compared to low-confidence errors – the typical hypercorrection effect. However, low-confidence errors to trick questions were just as likely to be corrected as high-confidence errors. Most surprisingly, we found that memory for the feedback and original responses, not confidence or surprise, were significant predictors of error correction. We conclude that for some types of material, there is an effortful process of elaboration and problem solving prior to making low-confidence errors that facilitates memory of corrective feedback.     

The effect of newly trained verbal and nonverbal labels for the cues in probabilistic category learning

Learning in a well-established paradigm of probabilistic category learning, the weather prediction task, has been assumed to be mediated by a variety of strategies reflecting explicit learning processes, such as hypothesis testing, when it is administered to young healthy participants. Higher categorization accuracy has been observed in the task when explicit processes are facilitated. We hypothesized that furnishing verbal labels for the cues would boost the formation, testing, and application of verbal rules, leading to higher categorization accuracy. We manipulated the availability of cue names by training separate groups of participants for three consecutive days to associate hard-to-name artificial auditory cues to pseudowords or to hard-to-name ideograms, or to associate stimulus intensity with colors; a fourth group remained unexposed to the cues. Verbal labels, cue individuation, and exposure to the stimulus set each had an additive effect on categorization performance in a subsequent 200-trial session of the weather prediction task using these auditory cues. This study suggests that cue nameability, when controlled for cue individuation and cue familiarity, has an effect on hypothesis-testing processes underlying category learning.     

Cross-task individual differences in error processing: Neural,electrophysiological, and genetic components

The error-related negativity (ERN) and error positivity (Pe) are electrophysiological markers of error processing thought to originate in the medial frontal cortex. Previous studies using probabilistic reinforcement showed that individuals who learn more from negative than from positive feedback (negative learners) had larger ERNs than did positive learners. These findings support the dopamine (DA) reinforcement-learning hypothesis of the ERN and associated computational models. However, it remains unclear (1) to what extent these effects generalize to tasks outside the restricted probabilistic reinforcement-learning domain and (2) whether there is a dopaminergic source of these effects. To address these issues, we tested subjects’ reinforcement-learning biases behaviorally and recorded EEG during an unrelated recognition memory experiment. Initial recognition responses were speeded, but the subjects were subsequently allowed to self-correct their responses. We found that negative learners, as assessed via probabilistic learning, had larger ERNs in the recognition memory task, suggestive of a common underlying enhanced error-processing mechanism. Negative learners also had enhanced Pes when selfcorrecting errors than did positive learners. Moreover, the ERN and Pe components contributed independently to negative learning. We also tested for a dopaminergic genetic basis of these ERP components. We analyzed the COMT val/met polymorphism, which has been linked to frontal DA levels. The COMT genotype affected Pe (but not ERN) magnitude; met/met homozygotes showed enhanced Pes to self-corrected errors, as compared with val carriers. These results are consistent with a role for the Pe and frontal monoamines in error awareness.     

Discounting of food,sex, and money

Discounting is a useful framework for understanding choice involving a range of delayed and probabilistic outcomes (e.g., money, food, drugs), but relatively few studies have examined how people discount other commodities (e.g., entertainment, sex). Using a novel discounting task, where the length of a line represented the value of an outcome and was adjusted using a staircase procedure, we replicated previous findings showing that individuals discount delayed and probabilistic outcomes in a manner well described by a hyperbola-like function. In addition, we found strong positive correlations between discounting rates of delayed, but not probabilistic, outcomes. This suggests that discounting of delayed outcomes may be relatively predictable across outcome types but that discounting of probabilistic outcomes may depend more on specific contexts. The generality of delay discounting and potential context dependence of probability discounting may provide important information regarding factors contributing to choice behavior.     

Causal learning from probabilistic events in 24‐month‐olds: an action measure

How do young children learn about causal structure in an uncertain and variable world? We tested whether they can use observed probabilistic information to solve causal learning problems. In two experiments, 24‐month‐olds observed an adult produce a probabilistic pattern of causal evidence. The toddlers then were given an opportunity to design their own intervention. In Experiment 1, toddlers saw one object bring about an effect with a higher probability than a second object. In Experiment 2, the frequency of the effect was held constant, though its probability differed. After observing the probabilistic evidence, toddlers in both experiments chose to act on the object that was more likely to produce the effect. The results demonstrate that toddlers can learn about cause and effect without trial‐and‐error or linguistic instruction on the task, simply by observing the probabilistic patterns of evidence resulting from the imperfect actions of other social agents. Such observational causal learning from probabilistic displays supports human children's rapid cultural learning.     

Choice patterns reveal qualitative individual differences among discounting of delayed gains,delayed losses,and probabilistic losses

A new probabilistic losses questionnaire as well as Kirby's delayed gains questionnaire and a previously developed delayed losses questionnaire were administered to a large online sample. Almost all participants showed the positive discounting choice pattern expected on the Kirby questionnaire, decreasing their choice of a delayed gain as time to its receipt increased. In contrast, approximately 15% of the participants showed negative discounting on the delayed losses questionnaire and/or the probabilistic losses questionnaire, decreasing their choice of an immediate loss as time to a delayed loss decreased and/or decreasing their choice of a certain loss as likelihood of the probabilistic loss increased. Mixture model analysis confirmed the existence of these negative discounting subgroups. The inconsistent findings observed in previous research involving delayed/probabilistic losses may be due to differences in the proportion of negative discounters who participated in previous studies. Further research is needed to determine how negative discounting of delayed and probabilistic losses manifests itself in everyday decisions. It should be noted that the presence of individuals who show atypical choice patterns when losses are involved may pose challenges for efforts to modify discounting in order to ameliorate behavioral problems, especially because many such problems concern choices that have negative consequences, often delayed and/or probabilistic.     

Strategies in probabilistic categorization: results from a new way of analyzing performance

The "Weather Prediction" task is a widely used task for investigating probabilistic category learning, in which various cues are probabilistically (but not perfectly) predictive of class membership. This means that a given combination of cues sometimes belongs to one class and sometimes to another. Prior studies showed that subjects can improve their performance with training, and that there is considerable individual variation in the strategies subjects use to approach this task. Here, we discuss a recently introduced analysis of probabilistic categorization, which attempts to identify the strategy followed by a participant. Monte Carlo simulations show that the analysis can, indeed, reliably identify such a strategy if it is used, and can identify switches from one strategy to another. Analysis of data from normal young adults shows that the fitted strategy can predict subsequent responses. Moreover, learning is shown to be highly nonlinear in probabilistic categorization. Analysis of performance of patients with dense memory impairments due to hippocampal damage shows that although these patients can change strategies, they are as likely to fall back to an inferior strategy as to move to more optimal ones.     

Naïve and Robust: Class‐Conditional Independence in Human Classification Learning

Humans excel in categorization. Yet from a computational standpoint, learning a novel probabilistic classification task involves severe computational challenges. The present paper investigates one way to address these challenges: assuming class‐conditional independence of features. This feature independence assumption simplifies the inference problem, allows for informed inferences about novel feature combinations, and performs robustly across different statistical environments. We designed a new Bayesian classification learning model (the dependence‐independence structure and category learning model, DISC‐LM) that incorporates varying degrees of prior belief in class‐conditional independence, learns whether or not independence holds, and adapts its behavior accordingly. Theoretical results from two simulation studies demonstrate that classification behavior can appear to start simple, yet adapt effectively to unexpected task structures. Two experiments—designed using optimal experimental design principles—were conducted with human learners. Classification decisions of the majority of participants were best accounted for by a version of the model with very high initial prior belief in class‐conditional independence, before adapting to the true environmental structure. Class‐conditional independence may be a strong and useful default assumption in category learning tasks.