Many of the problems studied in cognitive science are inductive problems, requiring people to evaluate hypotheses in the light of data. The key to solving these problems successfully is having the right inductive biases—assumptions about the world that make it possible to choose between hypotheses that are equally consistent with the observed data. This article explores a novel experimental method for identifying the biases that guide human inductive inferences. The idea behind this method is simple: This article uses the responses (...) produced by a participant on one trial to generate the stimuli that either they or another participant will see on the next. A formal analysis of this “iterated learning” procedure, based on the assumption that the learners are Bayesian agents, predicts that it should reveal the inductive biases of these learners, as expressed in a prior probability distribution over hypotheses. This article presents a series of experiments using stimuli based on a well-studied set of category structures, demonstrating that iterated learning can be used to reveal the inductive biases of human learners. (shrink)

We owe most scientific knowledge to methods of inquiry that are never formally analyzed. The analysis of behavior does not call for hypothetico-deductive methods. Statistics, taught in lieu of scientific method, is incompatible with major features of much laboratory research. Squeezing significance out of ambiguous data discourages the more promising step of scrapping the experiment and starting again. As a consequence, psychologists have taken flight from the laboratory. They have fled to Real People and the human interest of “real life,” (...) to Mathematical Models and the elegance of symbolic treatments, to the Inner Man and the explanatory preoccupation with inferred internal mechanisms, and to Laymanship and its appeal to “common sense.” An experimental analysis provides an alternative to these divertissements.The “theories” to which objection is raised here are not the basic assumptions essential to any scientific activity or statements that are not yet facts, but rather explanations which appeal to events taking place somewhere else, at some other level of observation, described in different terms, and measured, if at all, in different dimensions. Three types of learning theories satisfy this definition: physiological theories attempting to reduce behavior to events in the nervous system; mentalistic theories appealing to inferred inner events; and theories of the Conceptual Nervous System offered as explanatory models of behavior. It would be foolhardy to deny the achievements of such theories in the history of science. The question of whether they are necessary, however, has other implications.Experimental material in three areas illustrates the function of theory more concretely. Alternatives to behavior ratios, excitatory potentials, and so on demonstrate the utility of rate or probability of response as the basic datum in learning. Functional relations between behavior and environmental variables provide an account of why learning occurs. Activities such as preferring, choosing, discriminating, and matching can be dealt with solely in terms of behavior, without referring to processes in another dimensional system. The experiments are not offered as demonstrating that theories are not necessary but to suggest an alternative. Theory is possible in another sense. Beyond the collection of uniform relationships lies the need for a formal representation of the data reduced to a minimal number of terms. A theoretical construction may yield greater generality than any assemblage of facts; such a construction will not refer to another dimensional system. (shrink)

This paper shows how to define probability distributions over linguistically realistic syntactic structures in a way that permits us to define language learning and language comprehension as statistical problems. We demonstrate our approach using lexical‐functional grammar (LFG), but our approach generalizes to virtually any linguistic theory. Our probabilistic models are maximum entropy models. In this paper we concentrate on statistical inference procedures for learning the parameters that define these probability distributions. We point out some of the practical problems (...) that make straightforward ways of estimating these distributions infeasible, and develop a “pseudo‐likelihood” estimation procedure that overcomes some of these problems. This method raises interesting questions concerning the nature of the data available to a language learner and the modularity of language learning and processing. (shrink)

In this paper, we use an experimental design to compare the performance of elicitation rules for subjective beliefs. Contrary to previous works in which elicited beliefs are compared to an objective benchmark, we consider a purely subjective belief framework. The performance of different elicitation rules is assessed according to the accuracy of stated beliefs in predicting success. We measure this accuracy using two main factors: calibration and discrimination. For each of them, we propose two statistical indexes and we compare (...) the rules’ performances for each measurement. The matching probability method provides more accurate beliefs in terms of discrimination, while the quadratic scoring rule reduces overconfidence and the free rule, a simple rule with no incentives, which succeeds in eliciting accurate beliefs. Nevertheless, the matching probability appears to be the best mechanism for eliciting beliefs due to its performances in terms of calibration and discrimination, but also its ability to elicit consistent beliefs across measures and across tasks, as well as its empirical and theoretical properties. (shrink)

The Adaptive Toolbox framework specifies heuristics for choice and categorisation that search through cues in previously learned orders (Gigerenzer & Todd, 1999). We examined the learning of three cue parameters defining different orders: discrimination rate (DR) (the probability that a cue points to a unique choice), validity (the probability of correct choice given that a cue discriminates), and success (the probability of correct choice). Success orderings are identical to those by expected information gain (Klayman & Ha, (...) 1987). In two experiments, participants made choices in real-world environments with objective outcome criteria. Participant ratings indicated some appropriate parameter learning when the relevant cue parameter values were highly dispersed. Rated orders of cue validity and DR were less distinct than the objective orders—learning one parameter may be biased towards success by variation in the other. Success ratings capture the variation in validity and DR as well as participants' perception of these parameters. (shrink)

The development of the aquaculture industry in Canada has triggered a conflict of a scope never seen before. As stated in Young and Matthews’ The Aquaculture Controversy, this debate has “mushroomed over the past several decades to become one of the most bitter and stubborn face-offs over industrial development ever witnessed in Canada” (Young and Matthews in The aquaculture controversy in Canada. Activism, policy and contested science. UBC Press, Vancouver, p 3, 2010). It opposes a wide variety of actors: from (...) industrial investors, scientists, politicians and environmentalists to Native associations and communities, citizens groups and local stakeholders. The opposition is fierce between those in favor of a flourishing and modern aquaculture of industrial nature and those who fear the dreadful consequences of such an industry. In particular, the possible implementation of biotechnology innovations, such as a genetically modified salmon, has made this debate coextensive with the GMO debate, thus multiplying the opposition’s spectrum of arguments against the industry. Throughout the debate, Canadian environmental NGOs like the Suzuki Foundation and Greenpeace Canada have assumed leadership over the opposition to aquaculture development with certain success. Their participation in the debate features numerous ethical concerns related to environmental health, a respect for wilderness and local human communities. Such a position can be associated overall with a form of ecocentric ethics or concern. Nevertheless, a careful examination of the technical solutions proposed by these NGOs reveals their embeddedness in biocentrism. Through the example taken from the Canadian debate on aquaculture development, this paper aims to highlight the conceptual difficulty of enacting ecocentric ethical positions beyond formal arguments. Because ecocentrism implies a true paradigm shift, not only in mentality, but also in the way we conceive our technical interventions in nature, biocentrism remains an important practical method to enact ethical positions related to environmental concerns in public debates. (shrink)

This paper is concerned with learners who aim to learn patterns in infinite binary sequences: shown longer and longer initial segments of a binary sequence, they either attempt to predict whether the next bit will be a 0 or will be a 1 or they issue forecast probabilities for these events. Several variants of this problem are considered. In each case, a no-free-lunch result of the following form is established: the problem of learning is a formidably difficult one, in that (...) no matter what method is pursued, failure is incomparably more common that success; and difficult choices must be faced in choosing a method of learning, since no approach dominates all others in its range of success. In the simplest case, the comparison of the set of situations in which a method fails and the set of situations in which it succeeds is a matter of cardinality (countable vs. uncountable); in other cases, it is a topological matter (meagre vs. co-meagre) or a hybrid computational-topological matter (effectively meagre vs. effectively co-meagre). (shrink)

The figures visualizing single and combined classifiers coming from decision trees group and Bayesian parametric and nonparametric discriminant functions show the importance of diversity of bagging or boosting combined models and confirm some theoretical outcomes suggested by other authors. For the three medical sets examined, decision trees, as well as linear and quadratic discriminant functions are useful for bagging and boosting. Classifiers, which do not show an increasing tendency for resubstitution errors in subsequent boosting deterministic procedures loops, are not useful (...) for fusion, e.g. kernel discriminant function. For the success of resampling classifiers’ fusion, the compromise be- tween accuracy and diversity is needed. Diversity important in the success of boosting and bagging may be assessed by concordance of base classifiers with the learning vector. (shrink)

A modified version of the mere exposure effect paradigm was utilized in an implicit artificial grammar learning task in an attempt to develop a procedure that would be more sensitive in assesing nonconscious learning processes than the methods currently utilized within the field of implicit learning. Subjects were presented with stimuli generated from a finite-state artificial grammar and then had to either decide if novel items conformed to the rule structure of the grammar or rate the degree to which they (...) liked novel items. Because the latter of the task was more indirect of the two procedures, subjects′ availability to discriminate between well-formed and ill-formed items on this liking task was taken as a more sensitive piece of evidence of implicit learning, compared to performance on the rule conformity task. A subsequent test of subjects′ explicit knowledge of the rules of the grammar showed that subjects do have some conscious knowledge of the artificial grammar, but subjects making liking decisions exhibited significantly less conscious rule knowledge than subjects initially making rule-conformity judgements. The findings are discussed in relation to synergistic relationships among implicit perception, implicit memory, and implicit learning proceses. (shrink)

Necessity and sufficiency are the building blocks of all successful explanations. Yet despite their importance, these notions have been conceptually underdeveloped and inconsistently applied in explainable artificial intelligence (XAI), a fast-growing research area that is so far lacking in firm theoretical foundations. Building on work in logic, probability, and causality, we establish the central role of necessity and sufficiency in XAI, unifying seemingly disparate methods in a single formal framework. We provide a sound and complete algorithm for computing explanatory (...) factors with respect to a given context, and demonstrate its flexibility and competitive performance against state of the art alternatives on various tasks. (shrink)

AdaBoost has been successfully applied to a number of classification tasks, seemingly defying problems of overfitting. AdaBoost performs gradient descent in an error function with respect to the margin. This method concentrates on the patterns which are hardest to learn. However, this property of AdaBoost can be disadvantageous for noisy problems. Indeed, theoretical analysis has shown that the margin distribution plays a crucial role in understanding this phenomenon. Loosely speaking, some outliers should be tolerated if this has the benefit of (...) substantially increasing the margin on the remaining points. In this paper, we propose new noise robust boosting methods using the concepts of ν-Support Vector Classification and Arc-GV. These methods allow for the probability of a pre-specified fraction ν of points to lie in the margin area or even on the wrong side of the decision boundary. This algorithms can give a nicely interpretable way of controlling the trade-off between minimizing the training error and capacity. (shrink)

In order to build high-quality concrete, it is imperative to know the raw materials in advance. It is possible to accurately predict the quality of concrete and the amount of raw materials used using machine learning-enhanced methods. An automated process based on machine learning strategies is proposed in this paper for predicting the compressive strength of concrete. Fusion-learning-based optimization is used in the proposed approach to generate a strong learner by pooling support vector regression models. The SVR technique proposes an (...) optimization method for finding the kernel radial basis function parameters based on improving the innovative gunner algorithm. As a result of AIG's diverse solutions, local optima are effectively avoided. Therefore, the novelty of our research is that, in solving the uncertainty of predicted outputs based on integrated models, we use fusion-learning-based optimization to improve regression discrimination. We also collected a standard dataset to analyze the proposed algorithm, and subsequently, the dataset was designed from concrete laboratory tests on 244 samples, seven features, and three outputs. Different regression intensities are determined by correlation analysis of responses. Regression fusion is sufficiently accurate to estimate the number of desired outcomes examined based on the appropriate input data sample. The best quality concrete can be achieved with an error rate of less than 5%. (shrink)

Sequence learning is fundamental to a wide range of cognitive functions. Explaining how sequences—and the relations between the elements they comprise—are learned is a fundamental challenge to cognitive science. However, although hundreds of articles addressing this question are published each year, the actual learning mechanisms involved in the learning of sequences are rarely investigated. We present three experiments that seek to examine these mechanisms during a typing task. Experiments 1 and 2 tested learning during typing single letters on each trial. (...) Experiment 3 tested for “chunking” of these letters into “words.” The results of these experiments were used to examine the mechanisms that could best account for them, with a focus on two particular proposals: statistical transitional probability learning and discriminative error-driven learning. Experiments 1 and 2 showed that error-driven learning was a better predictor of response latencies than either n-gram frequencies or transitional probabilities. No evidence for chunking was found in Experiment 3, probably due to interspersing visual cues with the motor response. In addition, learning occurred across a greater distance in Experiment 1 than Experiment 2, suggesting that the greater predictability that comes with increased structure leads to greater learnability. These results shed new light on the mechanism responsible for sequence learning. Despite the widely held assumption that transitional probability learning is essential to this process, the present results suggest instead that the sequences are learned through a process of discriminative learning, involving prediction and feedback from prediction error. (shrink)

The article has several theses. First we propose that there is a new method of constitution-making today, the two-stage, post-sovereign one perfected in South Africa. Second, we admit the path-dependent nature, and difficult pre-conditions, of this method. Third, we maintain that even when the full method is unlikely in a given context, its legitimating principles nevertheless can play a role through international dissemination. We explore that possibility in the context of the projected comprehensive reform of Turkey, and the constitutional revolution (...) in Egypt. It is our belief that in these contexts one can learn both from successes of the new method and also from its failures typified by the Hungarian case that we briefly present. We are unfortunately not optimistic about the success of the new method especially where actors maintain their strong belief in the constituent power of the popular sovereign. This is likely to be the case in revolutions, but can happen in reform or even during the last state of the post-sovereign method itself. (shrink)

We report the results of an experiment in which human subjects were trained to perform a perceptual matching task. Subjects were asked to manipulate comparison objects until they matched target objects using the fewest manipulations possible. An unusual feature of the experimental task is that efficient performance requires an understanding of the hidden or latent causal structure governing the relationships between actions and perceptual outcomes. We use two benchmarks to evaluate the quality of subjects’ learning. One benchmark is based on (...) optimal performance as calculated by a dynamic programming procedure. The other is based on an adaptive computational agent that uses a reinforcement-learning method known as Q-learning to learn to perform the task. Our analyses suggest that subjects were successful learners. In particular, they learned to perform the perceptual matching task in a near-optimal manner (i.e., using a small number of manipulations) at the end of training. Subjects were able to achieve near-optimal performance because they learned, at least partially, the causal structure underlying the task. In addition, subjects’ performances were broadly consistent with those of model-based reinforcement-learning agents that built and used internal models of how their actions influenced the external environment. We hypothesize that people will achieve near-optimal performances on tasks requiring sequences of action—especially sensorimotor tasks with underlying latent causal structures—when they can detect the effects of their actions on the environment, and when they can represent and reason about these effects using an internal mental model. (shrink)

This doctoral dissertation investigates what influence indicative conditionals have on belief updating and how learning from conditionals may be modelled in a probabilistic framework. Because the problem is related to the interpretation of conditionals, we first assess different semantics of indicative conditionals. We propose that conditionals should be taken as primary concepts. This allows us to defend a claim that learning a conditional is equivalent to learning that the relevant conditional probability is 1. This implies that learning a conditional (...) can be modelled as learning a material conditional. We then turn to a few competing approaches to the problem of learning from a conditional -- an approach that models these situations with regards to changes in the explanatory status of the antecedent, as minimising probability distances between the prior and posterior probability distributions, and, particularly, an approach based on epistemic entrenchment. It is argued that although all approaches are important for their pioneering contributions, none of them ultimately provides a fully satisfactory solution. We then formulate a proposal according to which learning a conditional may be modelled as learning a material conditional in conjunction with other contextually inferred or observed information. We show that the view has an intuitive appeal on the grounds of pragmatics and empirical research in psychology of reasoning. We also show that the proposed approach successfully resolves all standard cases from the literature. The proposal is then extended to left-nested conditionals. To do so, we first inspect some preliminary empirical data about actual reasoning with nested conditionals and then show that the proposed approach also successfully models these cases. We then investigate how the approach may be extended to the so-called imperative conditionals. We introduce a tripartite typology of imperative conditionals and argue that although imperatives cannot be probabilistically modelled in any straight-forward way, at least some general insights of the proposed method help us understand the influence that the three different types of imperative conditionals have on listeners’ beliefs. We conclude with a summary and some open questions that should be addressed in future work (e.g. the influence of counterfactuals on belief updating). (shrink)

Objective: Tangent Space Mapping using the geometric structure of the covariance matrices is an effective method to recognize multiclass motor imagery. Compared with the traditional CSP method, the Riemann geometric method based on TSM takes into account the nonlinear information contained in the covariance matrix, and can extract more abundant and effective features. Moreover, the method is an unsupervised operation, which can reduce the time of feature extraction. However, EEG features induced by MI mental activities of different subjects are not (...) the same, so selection of subject-specific discriminative EEG frequency components play a vital role in the recognition of multiclass MI. In order to solve the problem, a discriminative and multi-scale filter bank tangent space mapping algorithm is proposed in this article to design the subject-specific Filter Bank so as to effectively recognize multiclass MI tasks. Methods: On the 4-class BCI competition IV-2a dataset, first, a non-parametric method of multivariate analysis of variance based on the sum of squared distances is used to select discriminative frequency bands for a subject; next, a multi-scale FB is generated according to the range of these frequency bands, and then decompose multi-channel EEG of the subject into multiple sub-bands combined with several time windows. Then TSM algorithm is used to estimate Riemannian tangent space features in each sub-band and finally a liner Support Vector Machines is used for classification. Main Results: The analysis results show that the proposed discriminative FB enhances the multi-scale TSM algorithm, improves the classification accuracy and reduces the execution time during training and testing. On the 4-class BCI competition IV-2a dataset, the average session to session classification accuracy of nine subjects reached 77.33 ± 12.3%. When the training time and the test time are similar, the average classification accuracy is 2.56% higher than the latest TSM method based on multi-scale filter bank analysis technology. When the classification accuracy is similar, the training speed is increased by more than three times, and the test speed is increased two times more. Compared with Supervised Fisher Geodesic Minimum Distance to the Mean, another new variant based on Riemann geometry classifier, the average accuracy is 3.36% higher, we also compared with the latest Deep Learning method, and the average accuracy of 10-fold cross validation improved by 2.58%. Conclusion: Research shows that the proposed DMFBTSM algorithm can improve the classification accuracy of MI tasks. Significance: Compared with the MFBTSM algorithm, the algorithm proposed in this article is expected to select frequency bands with good separability for specific subject to improve the classification accuracy of multiclass MI tasks and reduce the feature dimension to reduce training time and testing time. (shrink)

In this article we present symmetric diffusion networks, a family of networks that instantiate the principles of continuous, stochastic, adaptive and interactive propagation of information. Using methods of Markovion diffusion theory, we formalize the activation dynamics of these networks and then show that they can be trained to reproduce entire multivariate probability distributions on their outputs using the contrastive Hebbion learning rule (CHL). We show that CHL performs gradient descent on an error function that captures differences between desired and (...) obtained continuous multivariate probability distributions. This allows the learning algorithm to go beyond expected values of output units and to approximate complete probability distributions on continuous multivariate activation spaces. We argue that learning continuous distributions is an important task underlying a variety of real‐life situations that were beyond the scope of previous connectionist networks. Deterministic networks, like back propagation, cannot learn this task because they are limited to learning average values of independent output units. Previous stochastic connectionist networks could learn probability distributions but they were limited to discrete variables. Simulations show that symmetric diffusion networks can be trained with the CHL rule to approximate discrete and continuous probability distributions of various types. (shrink)