Generalized relevance learning vector quantization

We propose a new scheme for enlarging generalized learning vector quantization (GLVQ) with weighting factors for the input dimensions. The factors allow an appropriate scaling of the input dimensions according to their relevance. They are adapted automatically during training according to the specific classification task whereby training can be interpreted as stochastic gradient descent on an appropriate error function. This method leads to a more powerful classifier and to an adaptive metric with little extra cost compared to standard GLVQ. Moreover, the size of the weighting factors indicates the relevance of the input dimensions. This proposes a scheme for automatically pruning irrelevant input dimensions. The algorithm is verified on artificial data sets and the iris data from the UCI repository. Afterwards, the method is compared to several well known algorithms which determine the intrinsic data dimension on real world satellite image data.

Ecological and evolutionary validity: comments on Johnson-Laird, Legrenzi, Girotto, Legrenzi, and Caverni's (1999) mental-model theory of extensional reasoning

The mental-model account of naive probabilistic reasoning by P. N. Johnson-Laird, P. Legrenzi, V. Girotto, M. S. Legrenzi, and J.-P. Caverni (1999) provides an opportunity to clarify several similarities and differences between it and ecological rationality (frequentist) accounts. First, ambiguities in the meaning of Bayesian reasoning can lead to disagreements and inappropriate arguments. Second, 2 conflated effects of using natural frequencies are noticed but not actually tested separately because of an artificial dissociation of frequency representations and natural sampling. Third, similarities are noted between the subset principle and the principle of natural sampling. Finally, some potentially misleading portrayals of the role of evolutionary factors in psychology are corrected. Mental-model theory, rather than better explaining probabilistic reasoning, may be able to use frequency representations as a key element in clarifying its own ambiguous constructs.

Applications of the self-organising map to reinforcement learning

This article is concerned with the representation and generalisation of continuous action spaces in reinforcement learning (RL) problems. A model is proposed based on the self-organising map (SOM) of Kohonen [Self Organisation and Associative Memory, 1987] which allows either the one-to-one, many-to-one or one-to-many structure of the desired state-action mapping to be captured. Although presented here for tasks involving immediate reward, the approach is easily extended to delayed reward. We conclude that the SOM is a useful tool for providing real-time, on-line generalisation in RL problems in which the latent dimensionalities of the state and action spaces are small. Scalability issues are also discussed.

Pictorial versus textual information and the ratio-bias effect

The ratio-bias effect refers to the perception that a low probability event is more likely when presented in the form of a larger, e.g., 10-in-100, than smaller, e.g., 1-in-10, numerical ratio. This phenomenon has been used to help distinguish between rational-analytic versus heuristic-automatic ways of problem solving. In the current study, responding in accordance with the ratio bias was more prevalent when choices were presented pictorially as opposed to textually. Results are discussed with respect to how various factors, including mode of information presentation, may affect whether problems are solved via heuristic, automatic processes as opposed to effortful, analytic strategies.

A self-organising network that grows when required

The ability to grow extra nodes is a potentially useful facility for a self-organising neural network. A network that can add nodes into its map space can approximate the input space more accurately, and often more parsimoniously, than a network with predefined structure and size, such as the Self-Organising Map. In addition, a growing network can deal with dynamic input distributions. Most of the growing networks that have been proposed in the literature add new nodes to support the node that has accumulated the highest error during previous iterations or to support topological structures. This usually means that new nodes are added only when the number of iterations is an integer multiple of some pre-defined constant, A. This paper suggests a way in which the learning algorithm can add nodes whenever the network in its current state does not sufficiently match the input. In this way the network grows very quickly when new data is presented, but stops growing once the network has matched the data. This is particularly important when we consider dynamic data sets, where the distribution of inputs can change to a new regime after some time. We also demonstrate the preservation of neighbourhood relations in the data by the network. The new network is compared to an existing growing network, the Growing Neural Gas (GNG), on a artificial dataset, showing how the network deals with a change in input distribution after some time. Finally, the new network is applied to several novelty detection tasks and is compared with both the GNG and an unsupervised form of the Reduced Coulomb Energy network on a robotic inspection task and with a Support Vector Machine on two benchmark novelty detection tasks.

BYY harmony learning, structural RPCL, and topological self-organizing on mixture models

The Bayesian Ying-Yang (BYY) harmony learning acts as a general statistical learning framework, featured by not only new regularization techniques for parameter learning but also a new mechanism that implements model selection either automatically during parameter learning or via a new class of model selection criteria used after parameter learning. In this paper, further advances on BYY harmony learning by considering modular inner representations are presented in three parts. One consists of results on unsupervisedmixture models, ranging from Gaussian mixture based Mean Square Error (MSE) clustering, elliptic clustering, subspace clustering to NonGaussian mixture based clustering not only with each cluster represented via either Bernoulli-Gaussian mixtures or independent real factor models, but also with independent component analysis implicitly made on each cluster. The second consists of results on supervised mixture-of-experts (ME) models, including Gaussian ME, Radial Basis Function nets, and Kernel regressions. The third consists of two strategies for extending the above structural mixtures into self-organized topological maps. All these advances are introduced with details on three issues, namely, (a) adaptive learning algorithms, especially elliptic, subspace, and structural rival penalized competitive learning algorithms, with model selection made automatically during learning; (b) model selection criteria for being used after parameter learning, and (c) how these learning algorithms and criteria are obtained from typical special cases of BYY harmony learning.

The neural basis of human error processing: reinforcement learning, dopamine, and the error-related negativity

The authors present a unified account of 2 neural systems concerned with the development and expression of adaptive behaviors: a mesencephalic dopamine system for reinforcement learning and a "generic" error-processing system associated with the anterior cingulate cortex. The existence of the error-processing system has been inferred from the error-related negativity (ERN), a component of the event-related brain potential elicited when human participants commit errors in reaction-time tasks. The authors propose that the ERN is generated when a negative reinforcement learning signal is conveyed to the anterior cingulate cortex via the mesencephalic dopamine system and that this signal is used by the anterior cingulate cortex to modify performance on the task at hand. They provide support for this proposal using both computational modeling and psychophysiological experimentation.

Timing in pigeons: effects of the similarity between intertrial interval and gap in a timing signal

Previous research suggests that when a fixed interval is interrupted (known as the gap procedure), pigeons tend to reset memory and start timing from 0 after the gap. However, because the ambient conditions of the gap typically have been the same as during the intertrial interval (ITI), ambiguity may have resulted. In the present experiment, the authors found that when ambient conditions during the gap were similar to the ITI, pigeons tended to reset memory, but when ambient conditions during the gap were different from the ITI, pigeons tended to stop timing, retain the duration of the stimulus in memory, and add to that time when the stimulus reappeared. Thus, when the gap was unambiguous, pigeons timed accurately.

Anterior cingulate cortex and the Stroop task: neuropsychological evidence for topographic specificity

Neuroimaging studies have implicated the anterior cingulate cortex (ACC) in many aspects of attention and cognition. Major theories of ACC function have proposed a role in conflict monitoring, executive control, response selection, and general arousal. Although the ACC is often treated as a unitary structure, extensive evidence suggests it exhibits anatomical and functional specificity. ACC activity during the Stroop color word interference task has been of particular interest. The purpose of the present study was to determine whether two different ACC subregions are necessary for intact color naming performance in the Stroop task. One experiment utilized blocked trial and mixed trial designs to emulate neuroimaging studies and to compare interference and facilitation effects, respectively. A third variant manipulated the probabilities of congruent and incongruent trials to alter levels of interference and cognitive control, or engagement of strategic processes, on a block by block basis. Two patients with focal lesions in either right mid-caudal (patient D.L.) or left rostral to mid-dorsal ACC (patient R.N.) exhibited distinctive performance profiles in these three versions of the Stroop task, providing further support for topographic specificity of function within the human ACC. Contrary to predictions from some neuroimaging experiments, damage to right mid-caudal ACC was associated with normal levels of interference and accurate performance on incongruent trials. Instead, D.L. showed reduced levels of facilitation relative to controls. Further, interference was not modulated by the probability manipulation in D.L., suggesting equivalently high levels of cognitive control in both conditions. Conversely, damage to left mid-dorsal ACC resulted in consistently lower accuracy on incongruent trials, indicating deficits in maintaining task set and inhibiting the automatic response. These results can help to constrain interpretations of ACC activations in functional imaging experiments of the Stroop task.

Probabilistic classification learning in Tourette syndrome

Tourette syndrome (TS) is characterised by stereotyped involuntary movements, called tics. Some evidence suggests that structural and functional abnormalities of the basal ganglia may explain these motor symptoms. In this study, the probabilistic classification learning (PCL) test was used to evaluate basal ganglia functions in 10 children with less severe tics (Yale Global Tic Severity Scale (YGTSS) scores<30) and in 10 children with more severe symptoms (YGTSS score>30). In the PCL task, participants are asked to decide whether different combinations of four geometric forms (cues) predict rainy or sunny weather. Each cue is probabilistically related to a weather outcome, and feedback is provided after each decision. After completion of the probabilistic stimulus-response learning procedure, subjects received a transfer test to assess explicit knowledge about the cues. The children with TS exhibited impaired learning in the PCL task in comparison with the 20 healthy control subjects. This impairment was more pronounced in the TS patients with severe symptoms, and there was a significant negative relationship between the final classification performance and the YGTSS scores. The patients showed normal learning in the transfer test. These results suggest that the neostriatal habit learning system, which may play a central role in the acquisition of probabilistic associations, is dysfunctional in TS, especially in the case of more severe motor symptoms. The classification performance and the severity of tics were independent of the explicit knowledge obtained during the test.

Dissociating habit and recollection: evidence from Parkinson's disease, amnesia and focal lesion patients

We investigated the role played by the striatum and the medial temporal lobes (MTL) in memory performance by testing patients with Parkinson's disease (PD) and amnesia with Hay and Jacoby's habit-learning task [Journal of Experimental Psychology: Learning, Memory and Cognition 22 (1996) 1323]. Using equations from Jacoby's process-dissociation procedure [Journal of Memory and Language 30 (1991) 513], we were able to separate out the contribution of habit (automatic memory) and recollection (intentional memory) to performance within a single probability-learning paradigm. Amnesics showed the expected dissociation of impaired recollection and intact habit, highlighting the important role of the MTL in recollective processing. Mild PD patients did not perform differently than matched controls for habit or recollection, however, moderate PD patients were impaired in their ability to rely on habit and in their ability to recollect specific information. The performance of focal lesion patients further supported the interpretation that PD patients have a significant deficit in automatic, habit-learning due to striatal dysfunction while their deficit in recollection may arise from impoverished frontal lobe contributions.

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.

Representation facilitates reasoning: what natural frequencies are and what they are not

A good representation can be crucial for finding the solution to a problem. Gigerenzer and Hoffrage (Psychol. Rev. 102 (1995) 684; Psychol. Rev. 106 (1999) 425) have shown that representations in terms of natural frequencies, rather than conditional probabilities, facilitate the computation of a cause's probability (or frequency) given an effect--a problem that is usually referred to as Bayesian reasoning. They also have shown that normalized frequencies--which are not natural frequencies--do not lead to computational facilitation, and consequently, do not enhance people's performance. Here, we correct two misconceptions propagated in recent work (Cognition 77 (2000) 197; Cognition 78 (2001) 247; Psychol. Rev. 106 (1999) 62; Organ. Behav. Hum. Decision Process. 82 (2000) 217): normalized frequencies have been mistaken for natural frequencies and, as a consequence, "nested sets" and the "subset principle" have been proposed as new explanations. These new terms, however, are nothing more than vague labels for the basic properties of natural frequencies.

Chances and frequencies in probabilistic reasoning: rejoinder to Hoffrage, Gigerenzer, Krauss, and Martignon

Do individuals unfamiliar with probability and statistics need a specific type of data in order to draw correct inferences about uncertain events? Girotto and Gonzalez (Cognition 78 (2001) 247) showed that naive individuals solve frequency as well as probability problems, when they reason extensionally, in particular when probabilities are represented by numbers of chances. Hoffrage, Gigerenzer, Krauss, and Martignon (Cognition 84 (2002) 343) argued that numbers of chances are natural frequencies disguised as probabilities, though lacking the properties of true probabilities. They concluded that we failed to demonstrate that naive individuals can deal with true probabilities as opposed to natural frequencies. In this paper, we demonstrate that numbers of chances do represent probabilities, and that naive individuals do not confuse numbers of chances with frequencies. We conclude that there is no evidence for the claim that natural frequencies have a special cognitive status, and the evolutionary argument that the human mind is unable to deal with probabilities.

Dynamics and constraints in insight problem solving

This article reports 2 experiments that investigated performance on a novel insight problem, the 8-coin problem. The authors hypothesized that participants would make certain initial moves (strategic moves) that seemed to make progress according to the problem instructions but that nonetheless would guarantee failure to solve the problem. Experiment 1 manipulated the starting state of the problem and showed that overall solution rates were lower when such strategic moves were available. Experiment 2 showed that failure to capitalize on visual hints about the correct first move was also associated with the availability of strategic moves. The results are interpreted in terms of an information-processing framework previously applied to the 9-dot problem. The authors argue that in addition to the operation of inappropriate constraints, a full account of insight problem solving must incorporate a dynamic that steers solution-seeking activity toward the constraints.

Children's predictions of consistency in people's actions

Past research suggests that young children are often reluctant to generalize about people's behavior. Three experiments involving 102 4-5-year-olds, 84 7-8-year-olds, and 107 adults explored the conditions under which inductive inferences about people are made. There was an age-based increase in propensity to predict consistency in psychological/intentional causal relations. Children often predicted change; people would behave differently in the future than they did in the past. Younger children limited predictions of consistency to non-psychological contexts. Older children showed some appreciation of stable motivations (e.g. traits, preferences). The results are consistent with the hypothesis that children's theories of mind emphasize situational influences, with personal influences appearing in middle-childhood.

When is schematic knowledge used in source monitoring?

Source monitoring involves judgments regarding the origin of information (M. K. Johnson, S. Hashtroudi, & D. S. Lindsay, 1993). When participants cannot remember the source in a source-monitoring task, they may guess according to their prior schematic knowledge (U. J. Bayen, G. V. Nakamura, S. E. Dupuis, & C.-L. Yang, 2000). The present study aimed at specifying conditions under which schematic knowledge is used in source monitoring. The authors examined the time course of schema-based guesses with a response-signal technique (A. V. Reed, 1973), and multinomial models that separate memory and guessing bias. Use of schematic knowledge was observed only when asymptotic old-new recognition was low. The time course of schematic-knowledge retrieval followed an exponential growth function. Implications for theories of source monitoring are discussed.

Learning to vary and varying to learn

We compared two sources of behavior variability: decreased levels of reinforcement and reinforcement contingent on variability itself. In Experiment 1, four groups of rats were reinforced for different levels of response-sequence variability: one group was reinforced for low variability, two groups were reinforced for intermediate levels, and one group was reinforced for very high variability. All of the groups experienced three different reinforcement frequencies for meeting their respective variability contingencies. Results showed that reinforcement contingencies controlled response variability more than did reinforcement frequencies. Experiment 2 showed that only those animals concurrently reinforced for high variability acquired a difficult-to-learn sequence; animals reinforced for low variability learned little or not at all. Variability was therefore controlled mainly by reinforcement contingencies, and learning increased as a function of levels of baseline variability. Knowledge of these relationships may be helpful to those who attempt to condition operant responses.

Loci of signal probability effects and of the attentional blink bottleneck

To investigate the locus of signal probability effects and the influence of stimulus quality on this locus, the authors manipulated probability in Task 2 of a psychological refractory period (PRP) paradigm. The effect was additive with stimulus onset asynchrony (SOA) when the target was not masked but underadditive with decreasing SOA when the target was masked. Even with masking, however, a range of probabilities had effects additive with SOA. The results suggest loci of stimulus probability before the PRP bottleneck as well as at or after the bottleneck. A second issue addressed was the locus of interference in the attentional blink (AB). The AB was larger when the probability of the first of 2 targets was lower. The results lead to the conclusion that one cause of the AB effect is a locus at least as late as the PRP bottleneck.

Dissociation between priming and recognition in the expression of sequential knowledge

Exposure to a repeating sequence of target stimuli in a speeded localization task can support both priming of sequence-consistent responses and recognition of sequence components. Here, a task is introduced in which measures of priming and recognition are obtained concurrently, and it is demonstrated that priming of sequence-consistent responses occurs even when test stimuli are not recognized. The results show that sequence knowledge can be expressed in the absence of conscious recognition. However, we also show that this result is consistent with a simple model in which priming and recognition depend on exactly the same underlying memory strength variable.

Behavioral and electrocortical evidence of an interaction between probability and task metrics in movement preparation

Recent neurophysiological evidence suggests that cognitive factors shape neural activity in cortical areas such as parietal (area 5), premotor, and primary motor cortex. The implication of these findings is that behavioral signatures of cognitive factors and movement-specific factors should likewise be interdependent. The present study provides evidence of this interdependence in both behavioral (reaction time) and electrophysiological (P300) measures. Subjects performed a two-choice pointing task, in which the angular distance between the two required movement directions (task metrics) and the probability of the two responses was varied. In a control condition, a single reaction was required in response to both stimuli to test for the influence of stimulus metrics. Results from the pointing task showed a clear interaction between the metrics and effects of probability. When the potential targets were widely separated (120 degrees), stimulus probability influences reaction time and P300 amplitude in the classic fashion (longer reaction times and larger P300 amplitudes to less probable responses). When pointing to targets that were narrowly separated (20 degrees), probability had no effect: both rare and frequent targets were "functionally frequent." The same interaction was not observed in the control condition, indicating that metrics were primarily influencing movement preparation rather than stimulus processing. The results are consistent with the theoretical framework of dynamic field theory and demonstrate that metrics are an important factor that must be taken into account when assessing the processes associated with movement preparation.

Second-order backward blocking and unovershadowing in human causal learning

De Houwer and Beckers (in press, Experiment 1) recently demonstrated that ratings about the relation between a target cue T2 and an outcome are higher when training involves CT1+ and T1T2+ followed by C+ trials than when training involves CT1+ and T1T2+ followed by C- trials. We replicated this study but now explicitly asked participants to rate the causal status of the cues both before and after the C+ or C- trials. Results showed that causal ratings for T2 were significantly higher after C+ trials than before C+ trials and that T2 received significantly lower ratings after C- trials than before C- trials. The results thus provide the first evidence for higher-order unovershadowing and higher-order backward blocking. In addition, the ratings for T1 revealed that first-order backward blocking (i.e., decrease in ratings for T1 as the result of C+ trials) was stronger than first-order unovershadowing (i.e., increase in ratings for T1 as the result of C- trials).

The malleability of anchoring effects

Anchoring effects--the assimilation of a numeric estimate to a previously considered standard--are typically described as very robust and persistent. Based on the assumption that judgmental anchoring involves a hypothesis-testing process in which judges actively seek and generate judgment-relevant target knowledge, it was assumed that anchoring effects might at the same time be fairly malleable. Specifically, subtle influences that change the nature of the tested hypothesis are likely to affect the magnitude of anchoring. Using a procedural priming task, judges were induced to focus on similarities versus differences during a series of anchoring tasks. The results demonstrate that the magnitude of the obtained effect critically depended on this manipulation. In particular, a more pronounced anchoring assimilation effect resulted for judges with a similarity rather than a difference focus. Implications of these findings for models of anchoring as well as for the nature of the anchoring phenomenon are discussed.

Preparatory processes in the task-switching paradigm: evidence from the use of probability cues

The purpose of the investigations was to dissociate processes of task preparation from task execution in the task-switching paradigm. The basic assumption was that task repetitions have 2 advantages over task shifts: an activation advantage as a result of the execution of the same task type in the pretrial, and an expectation advantage, because participants, in general, implicitly expect a repetition. In Experiments 1-3, the authors explicitly manipulated expectancies by presenting cues that announced a shift and/or a repetition with probabilities of 1.00, .75, .50, or .25. Increasing latencies with decreasing probability for shifts and repetitions show that the expectation advantage can be equalized by preparation. However, the activation advantage represented by constant shift costs between tasks of the same probability is not penetrable by preparation. In Experiments 4 and 5, the authors found evidence that preparation involves activation of the expected task and inhibition of distracting tasks.

Age dissociates recency and lag recency effects in free recall

The temporal relations among word-list items exert a powerful influence on episodic memory retrieval. Two experiments were conducted with younger and older adults in which the age-related recall deficit was examined by using a decomposition method to the serial position curve, partitioning performance into (a) the probability of first recall, illustrating the recency effect, and (b) the conditional response probability, illustrating the lag recency effect (M. W. Howard & M. J. Kahana, 1999). Although the older adults initiated recall in the same manner in both immediate and delayed free recall, temporal proximity of study items (contiguity) exerted a much weaker influence on recall transitions in older adults. This finding suggests that an associative deficit may be an important contributor to older adults' well-known impairment in free recall.