From conditioning to category learning: An adaptive network model

Situating Human Sexual Conditioning

Conditioning is often thought of as a basic, automatic learning process that has limited applicability to higher-level human behavior. In addition, conditioning is seen as separable from, and even secondary to, "innate" processes. These ideas involve some misconceptions. The aim of this article is to provide a clearer, more refined sense of human sexual conditioning. After providing some background information and reviewing what is known from laboratory conditioning studies, human sexual conditioning is compared to sexual conditioning in nonhumans, to "innate" sexual responding, and to other types of human learning processes. Recommendations for moving forward in human sexual conditioning research are included.

Probabilistic inference under time pressure leads to a cortical-to-subcortical shift in decision evidence integration

Real-life decision-making often involves combining multiple probabilistic sources of information under finite time and cognitive resources. To mitigate these pressures, people "satisfice", foregoing a full evaluation of all available evidence to focus on a subset of cues that allow for fast and "good-enough" decisions. Although this form of decision-making likely mediates many of our everyday choices, very little is known about the way in which the neural encoding of cue information changes when we satisfice under time pressure. Here, we combined human functional magnetic resonance imaging (fMRI) with a probabilistic classification task to characterize neural substrates of multi-cue decision-making under low (1500 ms) and high (500 ms) time pressure. Using variational Bayesian inference, we analyzed participants' choices to track and quantify cue usage under each experimental condition, which was then applied to model the fMRI data. Under low time pressure, participants performed near-optimally, appropriately integrating all available cues to guide choices. Both cortical (prefrontal and parietal cortex) and subcortical (hippocampal and striatal) regions encoded individual cue weights, and activity linearly tracked trial-by-trial variations in the amount of evidence and decision uncertainty. Under increased time pressure, participants adaptively shifted to using a satisficing strategy by discounting the least informative cue in their decision process. This strategic change in decision-making was associated with an increased involvement of the dopaminergic midbrain, striatum, thalamus, and cerebellum in representing and integrating cue values. We conclude that satisficing the probabilistic inference process under time pressure leads to a cortical-to-subcortical shift in the neural drivers of decisions.

Measuring the relative contributions of rule-based and exemplar-based processes in judgment: Validation of a simple model

Judgments and decisions can rely on rules to integrate cue information or on the retrieval of similar exemplars from memory. Research on exemplar-based processes in judgment has discovered several task variables influencing the dominant mode of processing. This research often aggregates data across participants or classifies them as using either exemplar-based or cue-based processing. It has been argued for theoretical and empirical reasons that both kinds of processes might operate together or in parallel. Hence, a classification of strategies may be a severe oversimplification that also sacrifices statistical power to detect task effects. We present a simple measurement tool combining both processing modes. The simple model contains a mixture parameter quantifying the relative contribution of both kinds of processes in a judgment and decision task. In three experiments, we validate the measurement model by demonstrating that instructions and task variables affect the mixture parameter in predictable ways, both in memory-based and screen-based judgments.

Associative Learning of Social Value in Dynamic Groups

Although humans live in societies that regularly demand engaging with multiple people simultaneously, little is known about social learning in group settings. In two experiments, we combined a Pavlovian learning framework with dyadic economic games to test whether blocking mechanisms support value-based social learning in the gain (altruistic dictators) and loss (greedy robbers) domains. Subjects first learned about an altruistic dictator, who subsequently made altruistic splits collectively with a partner. Results revealed that because the presence of the dictator already predicted the outcome, subjects did not learn to associate value with the partner. This social blocking effect was not observed in the loss domain: A kind robber's partner, who could steal all the subjects' money but stole little, acquired highly positive value-which biased subjects' subsequent behavior. These findings reveal how Pavlovian mechanisms support efficient social learning, while also demonstrating that violations of social expectations can attenuate how readily these mechanisms are recruited.

Reward Draws the Eye, Uncertainty Holds the Eye: Associative Learning Modulates Distractor Interference in Visual Search

Stimuli in our sensory environment differ with respect to their physical salience but moreover may acquire motivational salience by association with reward. If we repeatedly observed that reward is available in the context of a particular cue but absent in the context of another cue the former typically attracts more attention than the latter. However, we also may encounter cues uncorrelated with reward. A cue with 50% reward contingency may induce an average reward expectancy but at the same time induces high reward uncertainty. In the current experiment we examined how both values, reward expectancy and uncertainty, affected overt attention. Two different colors were established as predictive cues for low reward and high reward respectively. A third color was followed by high reward on 50% of the trials and thus induced uncertainty. Colors then were introduced as distractors during search for a shape target, and we examined the relative potential of the color distractors to capture and hold the first fixation. We observed that capture frequency corresponded to reward expectancy while capture duration corresponded to uncertainty. The results may suggest that within trial reward expectancy is represented at an earlier time window than uncertainty.

Base-rate sensitivity through implicit learning

Two experiments assessed the contributions of implicit and explicit learning to base-rate sensitivity. Using a factorial design that included both implicit and explicit learning disruptions, we tested the hypothesis that implicit learning underlies base-rate sensitivity from experience (and that explicit learning contributes comparatively little). Participants learned to classify two categories of simple stimuli (bar graph heights) presented in a 3:1 base-rate ratio. Participants learned either from “observational” training to disrupt implicit learning or “response” training which supports implicit learning. Category label feedback on each trial was followed either immediately or after a 2.5 second delay by onset of a working memory task intended to disrupt explicit reasoning about category membership feedback. Decision criterion values were significantly larger following response training, suggesting that implicit learning underlies base-rate sensitivity. Disrupting explicit processing had no effect on base-rate learning as long as implicit learning was supported. These results suggest base-rate sensitivity develops from experience primarily through implicit learning, consistent with separate learning systems accounts of categorization.

Media multitasking and implicit learning

Media multitasking refers to the simultaneous use of different forms of media. Previous research comparing heavy media multitaskers and light media multitaskers suggests that heavy media multitaskers have a broader scope of attention. The present study explored whether these differences in attentional scope would lead to a greater degree of implicit learning for heavy media multitaskers. The study also examined whether media multitasking behaviour is associated with differences in visual working memory, and whether visual working memory differentially affects the ability to process contextual information. In addition to comparing extreme groups (heavy and light media multitaskers) the study included analysis of people who media multitask in moderation (intermediate media multitaskers). Ninety-four participants were divided into groups based on responses to the media use questionnaire, and completed the contextual cueing and n-back tasks. Results indicated that the speed at which implicit learning occurred was slower in heavy media multitaskers relative to both light and intermediate media multitaskers. There was no relationship between working memory performance and media multitasking group, and no relationship between working memory and implicit learning. There was also no evidence for superior performance of intermediate media multitaskers. A deficit in implicit learning observed in heavy media multitaskers is consistent with previous literature, which suggests that heavy media multitaskers perform more poorly than light media multitaskers in attentional tasks due to their wider attentional scope.

Relation between the Rational Model and the Context Model of Categorization

A formal proof is provided that Anderson's (1990) rational model of categorization generalizes the Medin and Schaffer (1978) context model. According to the context model, people represent categories by storing individual exemplars in memory. According to the rational model, people represent categories in terms of multiple exemplar-clusters or prototypes. In both models, a multiplicative rule is used to compute the similarity of an item to the underlying category representations. In certain special cases, each multiple prototype in the rational model corresponds to an individual exemplar, and in these cases the rational model reduces to the context model. Preliminary quantitative comparisons between the models are illustrated to test whether the multiple-prototype view adds significant explanatory power over the pure exemplar view.

Networks and Theories: The Place of Connectionism in Cognitive Science

This article considers how connectionist modeling can contribute to understanding of human cognition. J argue that connectionist networks should not be thought of as theories or simulations of theories, but may nevertheless contribute to the development of theories.

Attribution of Causality to Common and Distinctive Elements of Compound Stimuli

Given the task of diagnosing the source of a patient's allergic reaction, college students judged the causal efficacy of common (X) and distinctive (A and B) elements of compound stimuli: AX and BX. As the differential correlation of AX and BX with the occurrence and nonoccurrence of the allergic reaction rose from .00 to 1.00, ratings of the distinctive A and B elements diverged; most importantly, ratings of the common X element fell. These causal judgments of humans closely parallel the conditioned responses of animals in associative learning studies, and clearly disclose that stimuli compete with one another for control over behavior.

Stimulus Generalization and Representation in Adaptive Network Models of Category Learning

An exponential-decay relationship between the probability of generalization and psychological distance has received considerable support from studies of stimulus generalization ( Shepard, 1958 ) and categorization ( Nosofsky, 1984 ). It is shown here how an approximate exponential generalization gradient emerges from stimulus representation assumptions isomorphic to a special case of Shepard's (1987) theory of stimulus generalization in a “configuralcue” network model of human learning that represents stimulus patterns in terms of elementary features and pairwise conjunctions of features ( Gluck & Bower, 1988b ; Gluck, Bower, & Hee, 1989 ). The network model can be viewed as a combination of Shepard's theory and an associative learning rule derived from Rescorla and Wagner's (1972) theory of classical conditioning.

Probability matching in perceptrons: Effects of conditional dependence and linear nonseparability

Probability matching occurs when the behavior of an agent matches the likelihood of occurrence of events in the agent’s environment. For instance, when artificial neural networks match probability, the activity in their output unit equals the past probability of reward in the presence of a stimulus. Our previous research demonstrated that simple artificial neural networks (perceptrons, which consist of a set of input units directly connected to a single output unit) learn to match probability when presented different cues in isolation. The current paper extends this research by showing that perceptrons can match probabilities when presented simultaneous cues, with each cue signaling different reward likelihoods. In our first simulation, we presented up to four different cues simultaneously; the likelihood of reward signaled by the presence of one cue was independent of the likelihood of reward signaled by other cues. Perceptrons learned to match reward probabilities by treating each cue as an independent source of information about the likelihood of reward. In a second simulation, we violated the independence between cues by making some reward probabilities depend upon cue interactions. We did so by basing reward probabilities on a logical combination (AND or XOR) of two of the four possible cues. We also varied the size of the reward associated with the logical combination. We discovered that this latter manipulation was a much better predictor of perceptron performance than was the logical structure of the interaction between cues. This indicates that when perceptrons learn to match probabilities, they do so by assuming that each signal of a reward is independent of any other; the best predictor of perceptron performance is a quantitative measure of the independence of these input signals, and not the logical structure of the problem being learned.

Learned Irrelevance and Cue Competition Using an Eriksen Flanker Task

Many studies have examined competition between cues for learning. Research examining cue competition has used cues that predict the occurrence of an outcome, or, in some rare cases, competition between cues that predict the absence of an outcome (predicting that an outcome explicitly will not occur). Alternatively, learned irrelevance occurs when a cue lacks the ability to predict the occurrence or absence of an outcome. Using an Eriksen flanker task, the present study evaluated competition among cues that do not have predictive value, that is, competition for learning that an outcome is unpredictable. Subjects' inability to predict the occurrence of compatible and incompatible trials was manipulated by presenting cues that were uncorrelated with these trial types. Accuracy results showed competition between cues possessing a lack of predictive ability. The results are discussed in terms of propositional and associative theories of learning.

An Adaptive Connectionist Model of Cognitive Dissonance

This article proposes an adaptive connectionist model that implements an attributional account of cognitive dissonance. The model represents an attitude as the connection between the attitude object and behavioral-affective outcomes. Dissonance arises when circumstantial constraints induce a mismatch between the model's (mental) prediction and discrepant behavior or affect. Reduction of dissonance by attitude change is accomplished through long-lasting changes in the connection weights using the error-correcting delta learning algorithm. The model can explain both the typical effects predicted by dissonance theory as well as some atypical effects (i.e., reinforcement effect), using this principle of weight changes and by giving a prominent role to affective experiences. The model was implemented in a standard feedforward connectionist network. Computer simulations showed an adequate fit with several classical dissonance paradigms (inhibition, initiation, forced compliance, free choice, & misattribution), as well as novel studies that underscore the role of affect. A comparison with an earlier constraint satisfaction approach (Shultz & Lepper, 1996) indicates that the feedforward implementation provides a similar fit with these shortcomings of this previous model.

Dual-Process Models in Social and Cognitive Psychology: Conceptual Integration and Links to Underlying Memory Systems

Models postulating 2 distinct processing modes have been proposed in several topic areas within social and cognitive psychology. We advance a new conceptual model of the 2 processing modes. The structural basis of the new model is the idea, supported by psychological and neuropsychological evidence, that humans possess 2 memory systems. One system slowly learns general regularities, whereas the other can quickly form representations of unique or novel events. Associative retrieval or pattern completion in the slow-learning system elicited by a salient cue constitutes the effortless processing mode. The second processing mode is more conscious and effortful; it involves the intentional retrieval of explicit, symbolically represented rulesfrom either memory system and their use to guide processing. After presenting our model, we review existing dual-process models in several areas, emphasizing their similar assumptions of a quick, effortless processing mode that rests on well-learned prior associations and a second, more effortful processing mode that involves rule-based inferences and is employed only when people have both cognitive capacity and motivation. New insights and implications of the model for several topic areas are outlined.

Behavioral informatics: Dynamical models for measuring and assessing behaviors for precision interventions

Poor health-related behaviors represent a major challenge to healthcare due to their significant impact on chronic and acute diseases and their effect on the quality of life. Recent advances in technology have enabled an unprecedented opportunity to assess objectively, unobtrusively and continuously human behavior and have opened the possibility of optimizing individual-tailored, precision interventions within the framework of behavioral informatics. A key prerequisite for this optimization is the ability to assess and predict effects of interventions. This is potentially achievable with computational models of behavior and behavior change. In this paper we describe various approaches to computational modeling and describe a new hybrid model based on a dual process theoretical framework for behavior change. The model leverages cognitive learning theories and is shown to be consistent with mobile intervention data. We also illustrate how system-theoretic approaches can be used to assess the effect of coaching and participants' health behaviors.

How One Cause Discounts or Augments Another: A Connectionist Account of Causal Competition

The authors investigated the degree of discounting and augmentation of a target cause by an alternative cause given a varying number of observations on the alternative cause while holding its degree of covariation constant. Two experiments showed that more observations of the alternative cause resulted in greater discounting or augmentation of a target cause. This sample size effect cannot be explained by current attribution theories based on statistical notions or belief updating but can be accounted for by a connectionist framework. In addition, the authors found that the sample size effect was stronger when the information was presented in a sequential trial-by-trial format as opposed to a summarized format but found no effect of information order. Possible extensions of statistical models with confidence weights that take account of sample size were considered and simulated but none accommodated the data as well as connectionist models.

Implicit learning mediates base rate acquisition in perceptual categorization

We explored the possibility, suggested by Koehler (Behavioral and Brain Sciences, 19, 1-53, 1996; also Spellman Behavioral and Brain Sciences, 19, 38, 1996), that implicit learning mediates the influence of base-rates on category knowledge acquired through direct experience. In two experiments, participants learned simple perceptual categories with unequal base-rates (i.e., presentation frequency). In Experiment 1, participants received either response training or observational training. In Experiment 2, participants received response training with either immediate or delayed feedback. In previous studies, observational training and delayed feedback training have been shown to disrupt implicit learning. We found that base-rate influence was weaker in these conditions when category discriminability was low (i.e., when category membership was difficult to determine). This conclusion was based on signal detection β values as well as decision-bound modeling results. Because these disruptions to implicit learning attenuate the base-rate effect, we conclude that implicit learning does indeed underlie the influence of base-rates learned through direct experience. This suggests that the implicit learning system postulated by the COVIS theory of categorization (Ashby, Alfonso-Reese, Turken, & Waldron Psychological Review, 105, 442-481, 1998) may be involved in developing sensitivity to category base-rates.

Modeling Probabilistic Categorization Data: Exemplar Memory and Connectionist Nets

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

Competition Between Outcomes

In both Pavlovian conditioning and human causal judgment, competition between cues is well known to occur when multiple cues are presented in compound and followed by an outcome. More questionable is the occurrence of competition between outcomes when a single cue is followed by multiple outcomes presented in compound. In the experiment reported here, we demonstrated blocking (a type of stimulus competition) between outcomes. When the cue predicted one outcome, its ability to predict a second outcome that was presented in compound with the first outcome was reduced. The procedure minimized the likelihood that the observed competition between outcomes arose from selective attention. The competition between outcomes that we observed is problematic for contemporary theories of learning.

Cue Competition Effects: Empirical Tests of Adaptive Network Learning Models

The ability to predict future consequences on the basis of previous experience with the current set of environmental cues is one of the most fundamental of all cognitive processes. This study investigated how the validity of one cue influences the effectiveness of another cue for predicting a criterion. The results demonstrate a cue competition effect—increasing the validity of one cue decreased the effectiveness of another cue in a linear prediction task, even though the two cues were statistically independent.

Cue Competition Effects: Theoretical Implications for Adaptive Network Learning Models

A feature-free method for testing adaptive network learning models is presented. The test is based on a property called the mean matching law, a. property shared by many adaptive network models of learning. As an application of this method, we prove that cue competition effects obtained with statistically independent cues cannot be explained by many previous adaptive network learning models, including those based on the delta learning rule. These results point to the need to incorporate competitive learning properties into adaptive network learning models.

Concepts, Categories, and Psychological Science

Why are psychologists not often vailed on for their expertise in connection with the larger problems of society? The nature of expertise is analyzed in the framework of a theory of concept formation and categorization. Expert-novice differences in approaches to problems, and. more generally, the nature of abstraction and judgment, are interpreted in terms of cognitive processes that operate by computations on an instance-based composite memory. Applied to “growing pains” of psychological science, the theory has implications for graduate training and the direction of research support.

Hippocampal Context Processing during Acquisition of a Predictive Learning Task Is Associated with Renewal in Extinction Recall

Renewal is defined as the recovery of an extinguished response if extinction and retrieval contexts differ. The context dependency of extinction, as demonstrated by renewal, has important implications for extinction-based therapies. Persons showing renewal (REN) exhibit higher hippocampal activation during extinction in associative learning than those without renewal (NOREN), demonstrating hippocampal context processing, and recruit ventromedial pFC in retrieval. Apart from these findings, brain processes generating renewal remain largely unknown. Conceivably, processing differences in task-relevant brain regions that ultimately lead to renewal may occur already in initial acquisition of associations. Therefore, in two fMRI studies, we investigated overall brain activation and hippocampal activation in REN and NOREN during acquisition of an associative learning task in response to presentation of a context alone or combined with a cue. Results of two studies demonstrated significant activation differences between the groups: In Study 1, a support vector machine classifier correctly assigned participants' brain activation patterns to REN and NOREN groups, respectively. In Study 2, REN and NOREN showed similar hippocampal involvement during context-only presentation, suggesting processing of novelty, whereas overall hippocampal activation to the context–cue compound, suggesting compound encoding, was higher in REN. Positive correlations between hippocampal activation and renewal level indicated more prominent hippocampal processing in REN. Results suggest that hippocampal processing of the context–cue compound rather than of context only during initial learning is related to a subsequent renewal effect. Presumably, REN participants use distinct encoding strategies during acquisition of context-related tasks, which reflect in their brain activation patterns and contribute to a renewal effect.

Neural substrates of cognitive biases during probabilistic inference

Decision making often requires simultaneously learning about and combining evidence from various sources of information. However, when making inferences from these sources, humans show systematic biases that are often attributed to heuristics or limitations in cognitive processes. Here we use a combination of experimental and modelling approaches to reveal neural substrates of probabilistic inference and corresponding biases. We find systematic deviations from normative accounts of inference when alternative options are not equally rewarding; subjects' choice behaviour is biased towards the more rewarding option, whereas their inferences about individual cues show the opposite bias. Moreover, inference bias about combinations of cues depends on the number of cues. Using a biophysically plausible model, we link these biases to synaptic plasticity mechanisms modulated by reward expectation and attention. We demonstrate that inference relies on direct estimation of posteriors, not on combination of likelihoods and prior. Our work reveals novel mechanisms underlying cognitive biases and contributions of interactions between reward-dependent learning, decision making and attention to high-level reasoning.

Humans are often biased in estimating the precise influence of probabilistic events on their decisions. Here, Khorsand and colleagues report a behavioural task that produces these biases in inference and describe a biophysically-plausible model that captures these behavioural deviations from optimal decision making.

Cue Competition Effects in Human Causal Learning

Five experiments involving human causal learning were conducted to compare the cue competition effects known as blocking and unovershadowing, in proactive and retroactive instantiations. Experiment 1 demonstrated reliable proactive blocking and unovershadowing but only retroactive unovershadowing. Experiment 2 replicated the same pattern and showed that the retroactive unovershadowing that was observed was interfered with by a secondary memory task that had no demonstrable effect on either proactive unovershadowing or blocking. Experiments 3a, 3b, and 3c demonstrated that retroactive unovershadowing was accompanied by an inflated memory effect not accompanying proactive unovershadowing. The differential pattern of proactive versus retroactive cue competition effects is discussed in relationship to amenable associative and inferential processing possibilities.

Biases in probabilistic category learning in relation to social anxiety

Instrumental learning paradigms are rarely employed to investigate the mechanisms underlying acquired fear responses in social anxiety. Here, we adapted a probabilistic category learning paradigm to assess information processing biases as a function of the degree of social anxiety traits in a sample of healthy individuals without a diagnosis of social phobia. Participants were presented with three pairs of neutral faces with differing probabilistic accuracy contingencies (A/B: 80/20, C/D: 70/30, E/F: 60/40). Upon making their choice, negative and positive feedback was conveyed using angry and happy faces, respectively. The highly socially anxious group showed a strong tendency to be more accurate at learning the probability contingency associated with the most ambiguous stimulus pair (E/F: 60/40). Moreover, when pairing the most positively reinforced stimulus or the most negatively reinforced stimulus with all the other stimuli in a test phase, the highly socially anxious group avoided the most negatively reinforced stimulus significantly more than the control group. The results are discussed with reference to avoidance learning and hypersensitivity to negative socially evaluative information associated with social anxiety.

Novelty and Inductive Generalization in Human Reinforcement Learning

In reinforcement learning (RL), a decision maker searching for the most rewarding option is often faced with the question: What is the value of an option that has never been tried before? One way to frame this question is as an inductive problem: How can I generalize my previous experience with one set of options to a novel option? We show how hierarchical Bayesian inference can be used to solve this problem, and we describe an equivalence between the Bayesian model and temporal difference learning algorithms that have been proposed as models of RL in humans and animals. According to our view, the search for the best option is guided by abstract knowledge about the relationships between different options in an environment, resulting in greater search efficiency compared to traditional RL algorithms previously applied to human cognition. In two behavioral experiments, we test several predictions of our model, providing evidence that humans learn and exploit structured inductive knowledge to make predictions about novel options. In light of this model, we suggest a new interpretation of dopaminergic responses to novelty.

Processing of action- but not stimulus-related prediction errors differs between active and observational feedback learning

Learning of stimulus-response-outcome associations is driven by outcome prediction errors (PEs). Previous studies have shown larger PE-dependent activity in the striatum for learning from own as compared to observed actions and the following outcomes despite comparable learning rates. We hypothesised that this finding relates primarily to a stronger integration of action and outcome information in active learners. Using functional magnetic resonance imaging, we investigated brain activations related to action-dependent PEs, reflecting the deviation between action values and obtained outcomes, and action-independent PEs, reflecting the deviation between subjective values of response-preceding cues and obtained outcomes. To this end, 16 active and 15 observational learners engaged in a probabilistic learning card-guessing paradigm. On each trial, active learners saw one out of five cues and pressed either a left or right response button to receive feedback (monetary win or loss). Each observational learner observed exactly those cues, responses and outcomes of one active learner. Learning performance was assessed in active test trials without feedback and did not differ between groups. For both types of PEs, activations were found in the globus pallidus, putamen, cerebellum, and insula in active learners. However, only for action-dependent PEs, activations in these structures and the anterior cingulate were increased in active relative to observational learners. Thus, PE-related activity in the reward system is not generally enhanced in active relative to observational learning but only for action-dependent PEs. For the cerebellum, additional activations were found across groups for cue-related uncertainty, thereby emphasising the cerebellum's role in stimulus-outcome learning.

Mechanisms of object recognition: what we have learned from pigeons

Behavioral studies of object recognition in pigeons have been conducted for 50 years, yielding a large body of data. Recent work has been directed toward synthesizing this evidence and understanding the visual, associative, and cognitive mechanisms that are involved. The outcome is that pigeons are likely to be the non-primate species for which the computational mechanisms of object recognition are best understood. Here, we review this research and suggest that a core set of mechanisms for object recognition might be present in all vertebrates, including pigeons and people, making pigeons an excellent candidate model to study the neural mechanisms of object recognition. Behavioral and computational evidence suggests that error-driven learning participates in object category learning by pigeons and people, and recent neuroscientific research suggests that the basal ganglia, which are homologous in these species, may implement error-driven learning of stimulus-response associations. Furthermore, learning of abstract category representations can be observed in pigeons and other vertebrates. Finally, there is evidence that feedforward visual processing, a central mechanism in models of object recognition in the primate ventral stream, plays a role in object recognition by pigeons. We also highlight differences between pigeons and people in object recognition abilities, and propose candidate adaptive specializations which may explain them, such as holistic face processing and rule-based category learning in primates. From a modern comparative perspective, such specializations are to be expected regardless of the model species under study. The fact that we have a good idea of which aspects of object recognition differ in people and pigeons should be seen as an advantage over other animal models. From this perspective, we suggest that there is much to learn about human object recognition from studying the "simple" brains of pigeons.

Changes in GABA and glutamate concentrations during memory tasks in patients with Parkinson's disease undergoing DBS surgery

Until now direct neurochemical measurements during memory tasks have not been accomplished in the human basal ganglia. It has been proposed, based on both functional imaging studies and psychometric testing in normal subjects and in patients with Parkinson’s disease (PD), that the basal ganglia is responsible for the performance of feedback-contingent implicit memory tasks. To measure neurotransmitters, we used in vivo microdialysis during deep brain stimulation (DBS) surgery. We show in the right subthalamic nucleus (STN) of patients with PD a task-dependent change in the concentrations of glutamate and GABA during an implicit memory task relative to baseline, while no difference was found between declarative memory tasks. The five patients studied had a significant decrease in the percent concentration of GABA and glutamate during the performance of the weather prediction task (WPT). We hypothesize, based on current models of basal ganglia function, that this decrease in the concentration is consistent with expected dysfunction in basal ganglia networks in patients with PD.

The role of partial knowledge in statistical word learning

A critical question about the nature of human learning is whether it is an all-or-none or a gradual, accumulative process. Associative and statistical theories of word learning rely critically on the later assumption: that the process of learning a word's meaning unfolds over time. That is, learning the correct referent for a word involves the accumulation of partial knowledge across multiple instances. Some theories also make an even stronger claim: partial knowledge of one word-object mapping can speed up the acquisition of other word-object mappings. We present three experiments that test and verify these claims by exposing learners to two consecutive blocks of cross-situational learning, in which half of the words and objects in the second block were those that participants failed to learn in Block 1. In line with an accumulative account, Re-exposure to these mis-mapped items accelerated the acquisition of both previously experienced mappings and wholly new word-object mappings. But how does partial knowledge of some words speed the acquisition of others? We consider two hypotheses. First, partial knowledge of a word could reduce the amount of information required for it to reach threshold, and the supra-threshold mapping could subsequently aid in the acquisition of new mappings. Alternatively, partial knowledge of a word's meaning could be useful for disambiguating the meanings of other words even before the threshold of learning is reached. We construct and compare computational models embodying each of these hypotheses and show that the latter provides a better explanation of the empirical data.

Costs and benefits of automatization in category learning of ill-defined rules

Learning ill-defined categories (such as the structure of Medin & Schaffer, 1978) involves multiple learning systems and different corresponding category representations, which are difficult to detect. Application of latent Markov analysis allows detection and investigation of such multiple latent category representations in a statistically robust way, isolating low performers and quantifying shifts between latent strategies. We reanalyzed data from three experiments presented in Johansen and Palmeri (2002), which comprised prolonged training of ill-defined categories, with the aim of studying the changing interactions between underlying learning systems. Our results broadly confirm the original conclusion that, in most participants, learning involved a shift from a rule-based to an exemplar-based strategy. Separate analyses of latent strategies revealed that (a) shifts from a rule-based to an exemplar-based strategy resulted in an initial decrease of speed and an increase of accuracy; (b) exemplar-based strategies followed a power law of learning, indicating automatization once an exemplar-based strategy was used; (c) rule-based strategies changed from using pure rules to rules-plus-exceptions, which appeared as a dual processes as indicated by the accuracy and response-time profiles. Results suggest an additional pathway of learning ill-defined categories, namely involving a shift from a simple rule to a complex rule after which this complex rule is automatized as an exemplar-based strategy.

The widespread influence of the Rescorla-Wagner model

The theory of Pavlovian conditioning presented by Robert Rescorla and Allan Wagner in 1972 (the Rescorla-Wagner model) has been enormously important in animal learning research. It also has been applied in a variety of areas other than animal learning. We summarize the contribution of the Rescorla-Wagner model to research in verbal learning, social psychology, human category learning, human judgments of correlational relationships, transitive inference, color aftereffects, and physiological regulation. We conclude that there have been few models in experimental psychology as influential as the Rescorla-Wagner model.

Contingency Bias in Probability Judgement May Arise from Ambiguity regarding Additional Causes

In laboratory contingency learning tasks, people usually give accurate estimates of the degree of contingency between a cue and an outcome. However, if they are asked to estimate the probability of the outcome in the presence of the cue, they tend to be biased by the probability of the outcome in the absence of the cue. This bias is often attributed to an automatic contingency detection mechanism, which is said to act via an excitatory associative link to activate the outcome representation at the time of testing. We conducted 3 experiments to test alternative accounts of contingency bias. Participants were exposed to the same outcome probability in the presence of the cue, but different outcome probabilities in the absence of the cue. Phrasing the test question in terms of frequency rather than probability and clarifying the test instructions reduced but did not eliminate contingency bias. However, removal of ambiguity regarding the presence of additional causes during the test phase did eliminate contingency bias. We conclude that contingency bias may be due to ambiguity in the test question, and therefore it does not require postulation of a separate associative link-based mechanism.