Existing Learning Models Research Articles

Learning Through Noticing: Theory and Experimental Evidence in Farming

Existing learning models attribute failures to learn to a lack of data. We model a different barrier. Given the large number of dimensions one could focus on when using a technology, people may fail to learn because they failed to notice important features of the data they possess. We conduct a field experiment with seaweed farmers to test a model of “learning through noticing”. We find evidence of a failure to notice: On some dimensions, farmers do not even know the value of their own input. Interestingly, trials show that these dimensions are the ones that farmers fail to optimize. Furthermore, consistent with the model, we find that simply having access to the experimental data does not induce learning. Instead, farmers change behavior only when presented with summaries that highlight the overlooked dimensions. We also draw out the implications of learning through noticing for technology adoption, agricultural extension, and the meaning of human capital.

Learning Models: An Assessment of Progress, Challenges and New Developments

Learning models extend the traditional discrete choice framework by postulating that consumers have incomplete information about product attributes, and that they learn about these attributes over time. In this survey we describe the literature on learning models that has developed over the past 20 years, using the model of Erdem and Keane (1996) as a unifying framework. We described how subsequent work has extended their modeling framework, and applied learning models to a wide range of different products and markets. We argue that learning models have contributed greatly to our understanding of consumer behavior, in particular in enhancing our understanding of brand loyalty and long run advertising effects. We also discuss the limitations of existing learning models and discuss potential extensions. One key challenge is to disentangle learning as a source of dynamics from other key mechanisms that may generate choice dynamics (inventories, habit persistence, etc.). Another is to enhance identification of learning models by collecting and utilizing direct measures of signals, perceptions and expectations.

Observations on the economic manufacture quantity model with learning and forgetting

AbstractThe economic order (manufacture) quantity model with learning and forgetting has attracted the attention of researchers since the 1960s. All the models developed have limitations, of which two will be addressed in this article. First, the models found in the literature do not address the problem of when the learning exponent b approaches or exceeds the value 1, where these models are mathematically invalid for the special cases of b=1 and 2, and not investigated for the cases 1<b<2 and b>2. Second, the models found in the literature assume that the holding cost per unit is fixed even though the unit production cost is decreasing because of learning. This limitation would suggest producing in smaller lot‐sizes than their optimal values. Mathematical models are developed, and existing learning models are revised to overcome these limitations. Numerical examples are presented with results discussed.

Interactive Learning in Transportation Networks with Uncertainty, Bounded Rationality, and Strategic Choice Behavior

Existing learning and adaptive behavior models are typically derived from reinforcement learning from actual experiences in dynamic non-stationary environments. When the dynamics involve uncertainty with respect to the choice behavior of other travelers, travelers might make particular choice options by taking into account their expectations about the behavior of other travelers. Consequently, they will learn not only from their own experiences but also from the extent to which their conjectures about the behavior of other travelers are consistent with actual choices. A general model of interactive learning behavior that evolves toward equilibrium in such strategic situations is described. The properties of the model are examined by using numerical computer simulations. The results of the simulations support the face validity of the formulated model.

Interactive Learning in Transportation Networks with Uncertainty, Bounded Rationality, and Strategic Choice Behavior: Quantal Response Model

Existing learning and adaptive behavior models are typically derived from reinforcement learning from actual experiences in dynamic non-stationary environments. When the dynamics involve uncertainty with respect to the choice behavior of other travelers, travelers might make particular choice options by taking into account their expectations about the behavior of other travelers. Consequently, they will learn not only from their own experiences but also from the extent to which their conjectures about the behavior of other travelers are consistent with actual choices. A general model of interactive learning behavior that evolves toward equilibrium in such strategic situations is described. The properties of the model are examined by using numerical computer simulations. The results of the simulations support the face validity of the formulated model.

Broadening the Tests of Learning Models

For many years psychological studies of the learning process have used a simulated medical diagnosis task in which symptom configurations are probabilistically related to diseases. Participants are given a set of symptoms and asked to indicate which disease is present, and feedback is given on each trial. We enrich this standard laboratory task in four different ways. First, the symptoms have four possible values (low, medium low, medium high, and high) rather than just two. Second, symptom configurations are generated from an expanded factorial design rather than a simple factorial design. Third, subjects are asked to make a continuous judgment indicating their confidence in the diagnosis, rather than simply a binary judgment. Fourth, cumulated performance scores, payoffs, and the availability of a historical summary of the outcomes are varied in order to assess how these treatments modulate performance. These enrichments provide a broader data set and more challenging tests of the models. Using 123 subjects each in 480 trials, we compare five existing learning models plus several variants, including the well-known Bayesian, fuzzy logic, connectionist, exemplar, and ALCOVE models. We find that the subjects do learn to distinguish the symptom configurations, that subjects are quite heterogeneous in their response to the task, and that only a small part of the variation across subjects arises from the differences in treatments. The most striking finding is that the model that best predicts subjects' behavior is a simple Bayesian model with a single fitted parameter for prior precision to capture individual differences. We use rolling regression techniques to elucidate the behavior of this model over time and find some evidence of over-response to current stimuli.

Experience-Weighted Attraction Learning in Coordination Games: Probability Rules, Heterogeneity, and Time-Variation

In earlier research we proposed an “experience-weighted attraction (EWA) learning” model for predicting dynamic behavior in economic experiments on multiperson noncooperative normal-form games. We showed that EWA learning model fits significantly better than existing learning models (choice reinforcement and belief-based models) in several different classes of games. The econometric estimation in that research adopted a representative agent approach and assumed that learning parameters are stationary across periods of an experiment. In addition, we used the logit (exponential) probability response function to transform attraction of strategies into choice probability. This paper allows for nonstationary learning parameters, permits two “segments” of players with different parameter values in order to allow for some heterogeneity, and compares the power and logit probability response functions. These specifications are estimated using experimental data from weak-link and median-action coordination games. Results show that players are heterogeneous and that they adjust their learning parameters over time very slightly. Logit probability response functions never fit worse than power functions, and generally fit better.

Towards realistic theories of learning

In computational learning theory continuous efforts are made to formulate models of machine learning that are more realistic than previously available models. Two of the most popular models that have been recently proposed, Valiant’s PAC learning model and Angluin’s query learning model, can be thought of as refinements of preceding models such as Gold’s classic paradigm of identification in the limit, in which the question ofhow fast the learning can take place is emphasized. A considerable amount of results have been obtained within these two frameworks, resolving the learnability questions of many important classes of functions and languages. These two particular learning models are by no means comprehensive, and many important aspects of learning are not directly addressed in these models. Aiming towards more realistic theories of learning, many new models and extensions of existing learning models that attempt to formalize such aspects have been developed recently. In this paper, we will review some of these new extensions and models in computational learning theory, concentrating in particular on those proposed and studied by researchers at Theory NEC Laboratory RWCP, and their colleagues at other institutions.

AN EXPERIMENTAL INVESTIGATION OF A MATHEMATICAL LEARNING MODEL1, 2

ABSTRACTWhen mathematical learning models are applied to experimental data, frequently the adequacy of the model is assessed solely on the basis of the agreement between the model and data. Most of the existing learning models have been found to fit the data very well, in spite of the fact that many of the models are based on different underlying assumptions.The criterion of goodness of fit is therefore inadequate, and should be supplemented by another consideration such as the usefulness of the parameters. A scientifically useful model should contain parameters which could be used in prediction. If simple functional relationships are found between the parameters for a group of subjects on a set of problems, then these parameters could be used to predict the performance of these subjects on new problems or of new subjects on the original problems.The purpose of the present study is to determine the scientific usefulness of a special case of a model derived by Audley and Jonkheere (1956). This special case of the model is a direct mathematical formulation of Thorndike's Law of Effect and is isomorphic to a model proposed by Gulliksen (1934). The model has three parameters. One parameter (po) represents the initial probability of a correct response. The other two parameters are learning parameters. They represent the effect (α) of a reinforcement of a correct response on the strength of a correct response and the effect (δ) of nonreinforcement of an incorrect response on the strength of an incorrect response.The model was applied to data from two experiments in which cats were trained on visual discrimination problems. In the first experiment, six cats were trained on a series of seven problems. The best fitting parameter values were factor analyzed to determine if a simple relationship existed between the parameters. A two‐factor solution was obtained for the learning parameters, and a one‐factor solution was obtained for the initial performance parameter.In the second experiment, nine “split brain” cats were trained on a visual discrimination problem, first with one eye occluded then with the other eye occluded. Previous split brain research has demonstrated that under the conditions used in this experiment no interocular transfer occurs. The learning parameters (α and δ) that were obtained for the first and second training had no significant differences. The reliability of the learning parameters (α and δ) was very high (.992 and .952 respectively). The reliability of the initial probability parameter (po) was very low (.09), and the difference between the parameter for the first to the second training was significant at the .10 level.In both experiments, no significant differences were found between the model and the data. The model was therefore considered to fit the data well.The high reliability and presence of a two‐factor solution for the learning parameters (α and δ) indicate that these parameters follow simple functional transformations from one problem to another and from one cat to another. The presence of simple functional relationships indicates that the parameters obtained from the performance of a new cat on any one problem can give a good approximation of his performance on the other problems.Even though the initial performance parameters (po) could be accounted for by a one‐factor solution, the low reliability indicates that this parameter may not be useful as a predictor across cats or across problems. However, the initial performance parameter is useful within any single cat and problem, because it allows the differences in initial performances to be partialed out when the problem difficulty or cat performances are studied.