From conditioning to category learning: An adaptive network model.

We used adaptive network theory to extend the Rescorla-Wagner (1972) least mean squares (LMS) model of associative learning to phenomena of human learning and judgment. In three experiments subjects learned to categorize hypothetical patients with particular symptom patterns as having certain diseases. When one disease is far more likely than another, the model predicts that subjects will substantially overestimate the diagnosticity of the more valid symptom for the rare disease. The results of Experiments 1 and 2 provide clear support for this prediction in contradistinction to predictions from probability matching, exemplar retrieval, or simple prototype learning models. Experiment 3 contrasted the adaptive network model with one predicting pattern-probability matching when patients always had four symptoms (chosen from four opponent pairs) rather than the presence or absence of each of four symptoms, as in Experiment 1. The results again support the Rescorla-Wagner LMS learning rule as embedded within an adaptive network model. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

LMS rules and the inverse base-rate effect: Comment on Gluck and Bower (1988).

M. A. Gluck and G. H. Bower (see record 1989-00340-001) suggested that through the use of the Rescorla-Wagner learning rule, a connectionist network might be able to model the inverse base-rate phenomenon found by D. L. Medin and S. M. Edelson (see record 1988-31640-001). I prove that a network of the type that they proposed does not capture this effect. However, one can also prove that with additional assumptions about the encoding of features, the Rescorla-Wagner learning rule can be made to model the inverse base-rate effect. The importance of these assumptions and an outline of how they might be tested are then discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Learning about environmental geometry: An associative model.

K. Cheng (1986) suggested that learning the geometry of enclosing surfaces takes place in a geometric module blind to other spatial information. Failures to find blocking or overshadowing of geometry learning by features near a goal seem consistent with this view. The authors present an operant model in which learning spatial features competes with geometry learning, as in the Rescorla-Wagner model. Relative total associative strength of cues at a location determines choice of that location and thus the frequencies of reward paired with each cue. The model shows how competitive learning of local features and geometry can appear to result in potentiation, blocking, or independence, depending on enclosure shape and kind of features. The model reproduces numerous findings from dry arenas and water mazes. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The least mean squares network with information coding: A model of cue learning.

Information coding, a method for coding variables in the least mean squares (LMSs) network, is proposed as an associative model of human cue learning. According to this model, participants in cue learning experiments assume the linear hypothesis and implement this hypothesis by updating their estimate of cue–target correlation on each trial. Because absence of a stimulus may give information about correlation, this model codes an absent stimulus as a nonzero value under some conditions. Thus, Information Coding differs from R. A. Rescorla and A. R. Wagner's (1972) model, a version of the LMSs network in which absence is always coded as 0. In 5 experiments, participants estimated the likelihood that a hypothetical patient had a disease given the patient's symptoms. The results support Information Coding, whereas the Rescorla-Wagner model, G. B. Chapman's (see record 1992-04135-001) variant of the Rescorla-Wagner model, and A. B. Markman's (see record 1990-09047-001) coding of the LMSs network all failed to account for at least one of these results. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

An associative model of geometry learning: A modified choice rule.

In a recent article, the authors (Miller & Shettleworth, 2007; see record 2007-09968-001) showed how the apparently exceptional features of behavior in geometry learning ("reorientation") experiments can be modeled by assuming that geometric and other features at given locations in an arena are learned competitively as in the Rescorla-Wagner model and that the probability of visiting a location is proportional to the total associative strength of cues at that location relative to that of all relevant locations. Reinforced or unreinforced visits to locations drive changes in associative strengths. Dawson, Kelly, Spetch, and Dupuis (2008; see record 2008-09669-009) have correctly pointed out that at parameter values outside the ranges the authors used to simulate a body of real experiments, our equation for choice probabilities can give impossible and/or wildly fluctuating results. Here, the authors show that a simple modification of the choice rule eliminates this problem while retaining the transparent way in which the model relates spatial choice to competitive associative learning of cue values. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Selective associations and causality judgments: Presence of a strong causal factor may reduce judgments of a weaker one.

In 5 experiments, humans played video games in which 2 events or causes covaried with an outcome. In Exps 1 and 2, a highly correlated cause (a plane) of an outcome (success at traversing a minefield) reduced judgments of the strength of a weaker cause (camouflaging or painting a tank). In Exp 3, similar results were found when both causes were negatively correlated with the outcome. In Exp 4, strong positive or negative contingencies caused the Ss to reduce judgments of contingencies of the opposite polarity. These results can be accounted for by associative or connectionist models from animal learning such as the Rescorla-Wagner model. In Exp 5, this type of model was contrasted with a representational model in which Ss are claimed to monitor accurately the various contingencies but use a rule in which the presence of a strong contingency causes them to discount weaker contingencies. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Associative versus contingency accounts of category learning: Reply to Melz, Cheng, Holyoak, and Waldmann (1993).

E. R. Melz et al (see record 1994-24205-001) argue that the partial blocking of cue A that was previously reported (D. R. Shanks; see record 1991-26433-001) when Ss were presented with intermixed AB?→?1, B?→?1 category learning trials is not consistent with the associative Rescorla-Wagner (R-W; 1972) theory analysis that was offered, given that the theory predicts complete blocking at asymptote. However, this claim assumes that Ss were trained to asymptote in these experiments, and there is no reason to believe this was the case. Melz et al further argue that there has been no reported evidence of complete blocking in associative learning tasks, which is incorrect. It is shown that, on the contrary, there is abundant evidence of it. The R-W theory analysis of the results is therefore sound. The results reported were inconsistent with contingency theories as they are normally formulated. Melz et al propose a revised contingency theory which, they argue, can account for data from a range of learning tasks. In particular, they claim that their theory can accommodate the results. It is shown that the theory can be refuted… (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Component and pattern information in adaptive networks.

M. A. Gluck and G. H. Bower (see record 1989-00340-001) presented an adaptive network model of human classification in which associative weights are modified according to R. A. Rescorla and A. R. Wagner (1972) conditioning theory, a special case of the Widrow-Hoff/LMS learning rule. Presenting empirical data from a series of artificial medical classification tasks, we argued that the network model predicts results that were unanticipated, given prevailing alternative theories of category learning. We consider here some alternative interpretations of this data suggested by D. R. Shanks (see record 1990-27514-001) and argue that they are not sufficiently compelling when compared to the network model's treatment of the data from all the experiments presented in our earlier study. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Critique of the retrieval/deblurring assumptions of the theory of distributed associative memory.

Critiques the retrieval assumptions proposed by S. Lewandowsky and B. B. Murdock (see record 1989-14457-001) in their application of Murdock's theory of distributed associative memory (TODAM) to problems in serial-order recall. Two different methods for "deblurring" the products of retrieval are described, along with simulations appropriate for each. It is demonstrated that the deblurring assumptions used by Lewandowsky and Murdock in their simulation version of the model, although appropriate for some aspects of serial-anticipation learning, provide an inadequate account of general serial recall and, in fact, predict several trends that are inconsistent with known data. The analytic version of the model is considered and some additional problems are noted. Overall, the deblurring assumptions critically affect the performance of TODAM, and it is these assumptions rather than the model per se that determine a sizable proportion of its behavior. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Cumulative effects model: A response to Williams (1994).

The cumulative effects (CE) model explains free-operant choice by the ratio of total numbers of responses and reinforcements, a probability-like variable. B. A. Williams (see record 1995-08399-001) argues that the model is vulnerable to experiments that disprove melioration, a local probability model. The authors note critical differences between the nonlocal CE model and local probability models that allow the CE model to handle some data with which they are incompatible. All models are simplifications of reality; hence, a model's failures are as revealing as its successes. Williams suggests that simple models may need to be abandoned in favor of a "representational" account. The authors point out that representations must be both acquired and acted on. Acquisition requires processing of responses and reinforcers; action requires decision rules. Models are simply testable suggestions for what these rules and processes might be. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Evaluation of blocking and conditioned inhibition to a CS signaling a decrease in US intensity

Two experiments employing a conditioned emotional response procedure with rats evaluated the associative tendencies acquired by a target CS when a compound of that stimulus and another CS was reinforced by a low-intensity US whereas the latter CS alone was reinforced by a higher intensity US. Experiment 1 involved a blocking sequence in which the element and compound trials occurred in successive phases, followed by eventual testing of the responding to the target CS alone. Less evidence of excitatory conditioning was observed than in a comparison group which had the element and compound paired with the same low-intensity US. Experiment 2 included contemporaneous differential reinforcement of the element and compound CSs and periodic summation tests in which the target CS was compounded with another independently trained CS. Clear evidence was obtained of the development of inhibitory tendencies by the target CS. The results are consistent with the predictions of the Rescorla-Wagner model and are discussed in relation to the several determining assumptions of the model.     

On the whats and hows of retrieval in the acquisition of a simple skill.

Two general views on the role of memory in cognitive skills—an instance-based theory and an associative perspective—were compared with respect to their general assumptions about the information involved and the processes that operate on that information. Characteristics of memory information were examined in terms of predictions for transfer to various stimulus forms as a function of 2 types of learning conditions. Characteristics of memory processes were examined using a set of general process models. Results of 4 experiments indicate that (a) neither theoretical perspective was capable of accounting for all the observed transfer effects, indicating needed refinements to informational assumptions, and that (b) 1 class of process assumptions was consistently supported, whereas other classes were consistently contradicted, indicating a general set of process characteristics that can be used in further model development. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

What is learned in sequential learning? An associative model of reward magnitude serial-pattern learning.

A computational model of sequence learning is described that is based on pairwise associations and generalization. Simulations by the model predicted that rats should learn a long monotonic pattern of food quantities better than a nonmonotonic pattern, as predicted by rule-learning theory, and that they should learn a short nonmonotonic pattern with highly discriminable elements better than 1 with less discriminable elements, as predicted by interitem association theory. In 2 other studies, the model also simulated behavioral "rule generalization", "extrapolation", and associative transfer data motivated by both rule-learning and associative perspectives. Although these simulations do not rule out the possibility that rats can use rule induction to learn serial patterns, they show that a simple associative model can account for the classical behavioral studies implicating rule learning in reward magnitude serial-pattern learning. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Associative History Affects the Associative Change Undergone by Both Presented and Absent Cues in Human Causal Learning.

R. A. Rescorla (2000) noted that a number of influential theories of associative learning do not take the associative history of cues (i.e., the prior training that they have received) into account when calculating the associative change undergone by those cues. The authors tested this assumption in a human causal learning paradigm and found associative history to be an important determinant of the learning undergone by cues that are presented on a trial. Moreover, associative history was also found to influence the amount of retrospective revaluation undergone by absent cues. These findings conflict with models of causal learning in which the associative change undergone by an element of a cue compound is governed by a summed error term (e.g., R. A. Rescorla & A. R. Wagner, 1972). (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

A connectionist account of asymmetric category learning in early infancy.

Young infants show unexplained asymmetries in the exclusivity of categories formed on the basis of visually presented stimuli. A connectionist model is described that shows similar exclusivity asymmetries when categorizing the same stimuli presented to infants. The asymmetries can be explained in terms of an associative learning mechanism, distributed internal representations, and the statistics of the feature distributions in the stimuli. The model was used to explore the robustness of this asymmetry. The model predicts that the asymmetry will persist when a category is acquired in the presence of mixed category exemplars. An experiment with 3–4-month-olds showed that asymmetric exclusivity persisted in the presence of mixed-exemplar familiarization, thereby confirming the model's prediction. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

How do we tell an association from a rule? Comment on Sloman (1996).

S. A. Sloman's (see PA, Vol 83:21925) intriguing argument for separate associative and rule-based reasoning systems is unfortunately damaged by a certain amount of slack in the distinction he makes between these two posited mental mechanisms. The authors suggest that the distinction could be sharpened by overt reference to explicit models of associative and rule-based processing. They also point out that "simultaneous contradictory belief," which Sloman takes as evidence for separate associative and rule-based systems, need not be interpreted in this fashion. It may also signal a number of other things, including the presence of linguistic ambiguity (as in the Linda problem), competing lines of formal reasoning (as in the Wason selection task), and unclarified assumptions (as in the 3 doors problem). (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Seeing Versus Doing: Two Modes of Accessing Causal Knowledge.

The ability to derive predictions for the outcomes of potential actions from observational data is one of the hallmarks of true causal reasoning. We present four learning experiments with deterministic and probabilistic data showing that people indeed make different predictions from causal models, whose parameters were learned in a purely observational learning phase, depending on whether learners believe that an event within the model has been merely observed ("seeing") or was actively manipulated ("doing"). The predictions reflect sensitivity both to the structure of the causal models and to the size of their parameters. This competency is remarkable because the predictions for potential interventions were very different from the patterns that had actually been observed. Whereas associative and probabilistic theories fail, recent developments of causal Bayes net theories provide tools for modeling this competency. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Stimulus Generalization in Two Associative Learning Processes.

Recent studies involving nonlinear discrimination problems suggest that stimuli in human associative learning are represented configurally with nairow generalization, such that presentation of stimuli that are even slightly dissimilar to stored configurations weakly activate these configurations. The authors note that another well-known set of findings in human associative learning, cue-interaction phenomena, suggest relatively broad generalization. Three experiments show that current models of human associative learning, which try to model both nonlinear discrimination and cue interaction as the result of 1 process, fail because they cannot simultaneously account for narrow and broad generalization. Results suggest that human associative learning involves (a) an exemplar-based process with configural stimulus representation and narrow generalization and (b) an adaptive learning process characterized by broad generalization and cue interaction. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Making causal judgments from the proportion of confirming instances: The pCI rule.

It is proposed that causal judgments about contingency information are derived from the proportion of confirmatory instances (pCI) that are evaluated as confirmatory for the causal candidate. In 6 experiments, pCI values were manipulated independently of objective contingencies assessed by the ΔP rule. Significant effects of the pCI manipulations were found in all cases, but causal judgments did not vary significantly with objective contingencies when pCI was held constant. The experiments used a variety of stimulus presentation procedures and different dependent measures. The power PC theory, a weighted version of the ΔP rule, the Rescorla-Wagner associative learning model (R. A. Rescorla & A. R. Wagner, 1972), and the ΔD rule, which is the frequency-based version of the pCI rule, were unable to account for the significant effects of the pCI manipulations. These results are consistent with a general explanatory approach to causal judgment involving the evaluation of evidence and updating of beliefs with regard to causal hypotheses. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Is the number of trials a primary determinant of conditioned responding?

Acquisition of conditioned responding is thought to be determined by the number of pairings of a conditioned stimulus (CS) and an unconditioned stimulus (US). However, it is possible that acquisition is primarily determined not by the number of trials but rather by quantities that often correlate with the number of trials, such as cumulative intertrial interval (ITI) and the number of sessions. Four experiments examined whether the number of trials has an effect on acquisition of conditioned responding, once cumulative ITI and number of sessions are equated. Results of the experiments with rats and mice favor the hypothesis that over an eightfold range, variation in number of CS-US pairings has little effect. It is suggested that learning curves might more accurately be plotted across cumulative ITI or number of sessions, and not across number of trials. Results pose a challenge to trial-centered accounts of conditioning, as demonstrated by simulations of the Rescorla-Wagner model, a simplified version of Wagner's standard operating procedure model (SOP), and Stout & Miller's sometimes competing retrieval model (SOCR). A time-centered account, rate estimation theory (RET), predicts the main finding but has trouble with other aspects of the learning process more easily accommodated by trial-centered models. (PsycINFO Database Record (c) 2010 APA, all rights reserved)