Children can solve Bayesian problems: the role of representation in mental computation

Can children reason the Bayesian way? We argue that the answer to this question depends on how numbers are represented, because a representation can do part of the computation. We test, for the first time, whether Bayesian reasoning can be elicited in children by means of natural frequencies. We show that when information was presented to fourth, fifth, and sixth graders in terms of probabilities, their ability to estimate the Bayesian posterior probability was zero. Yet when the same information was presented in natural frequencies, Bayesian reasoning showed a steady increase from fourth to sixth grade, reaching an average level of 19, 39, and 53%, respectively, in two studies. Sixth graders' performance with natural frequencies matched the performance of adults with probabilities. But this general increase was accompanied by striking individual differences. More than half of the sixth graders solved most or all problems, whereas one third could not solve a single one. An analysis of the children's responses provides evidence for the use of three non-Bayesian strategies. These follow an overlapping wave model of development and continue to be observed in the minds of adults. More so than adults' probabilistic reasoning, children's reasoning depends on a proper representation of information.     

Self-representation: searching for a neural signature of self-consciousness

Human self-consciousness operates at different levels of complexity and at least comprises five different levels of representational processes. These five levels are nonconceptual representation, conceptual representation, sentential representation, meta-representation, and iterative meta-representation. These different levels of representation can be operationalized by taking a first-person-perspective that is involved in representational processes on different levels of complexity. We refer to experiments that operationalize a first-person-perspective on the level of conceptual and meta-representational self-consciousness. Interestingly, these experiments show converging evidence for a recruitment of medial cortical and parietal regions during taking a first-person-perspective, even when operating on different degrees of complexity. These data lend support for the speculative hypothesis, that there exist a neural signature for human self-consciousness that is recruited independent from the degree of representational complexity to be performed.     

First- and third-person approaches in implicit learning research

How do we find out whether someone is conscious of some information or not? A simple answer is “We just ask them”! However, things are not so simple. Here, we review recent developments in the use of subjective and objective methods in implicit learning research and discuss the highly complex methodological problems that their use raises in the domain.     

Semantic interpretation as computation in nonmonotonic logic: the real meaning of the suppression task

Interpretation is the process whereby a hearer reasons to an interpretation of a speaker's discourse. The hearer normally adopts a credulous attitude to the discourse, at least for the purposes of interpreting it. That is to say the hearer tries to accommodate the truth of all the speaker's utterances in deriving an intended model. We present a nonmonotonic logical model of this process which defines unique minimal preferred models and efficiently simulates a kind of closed-world reasoning of particular interest for human cognition. Byrne's "suppression" data (Byrne, 1989) are used to illustrate how variants on this logic can capture and motivate subtly different interpretative stances which different subjects adopt, thus indicating where more fine-grained empirical data are required to understand what subjects are doing in this task. We then show that this logical competence model can be implemented in spreading activation network models. A one pass process interprets the textual input by constructing a network which then computes minimal preferred models for (3-valued) valuations of the set of propositions of the text. The neural implementation distinguishes easy forward reasoning from more complex backward reasoning in a way that may be useful in explaining directionality in human reasoning.     

Uncovering the richness of the stimulus: structure dependence and indirect statistical evidence

The poverty of stimulus argument is one of the most controversial arguments in the study of language acquisition. Here we follow previous approaches challenging the assumption of impoverished primary linguistic data, focusing on the specific problem of auxiliary (AUX) fronting in complex polar interrogatives. We develop a series of corpus analyses of child-directed speech showing that there is indirect statistical information useful for correct auxiliary fronting in polar interrogatives and that such information is sufficient for distinguishing between grammatical and ungrammatical generalizations, even in the absence of direct evidence. We further show that there are simple learning devices, such as neural networks, capable of exploiting such statistical cues, producing a bias toward correct AUX questions when compared to their ungrammatical counterparts. The results suggest that the basic assumptions of the poverty of stimulus argument may need to be reappraised.     

A learning rule that explains how rewards teach attention

Many theories propose that top-down attentional signals control processing in sensory cortices by modulating neural activity. But who controls the controller? Here we investigate how a biologically plausible neural reinforcement learning scheme can create higher order representations and top-down attentional signals. The learning scheme trains neural networks using two factors that gate Hebbian plasticity: (1) an attentional feedback signal from the response-selection stage to earlier processing levels; and (2) a globally available neuromodulator that encodes the reward prediction error. We demonstrate how the neural network learns to direct attention to one of two coloured stimuli that are arranged in a rank-order. Like monkeys trained on this task, the network develops units that are tuned to the rank-order of the colours and it generalizes this newly learned rule to previously unseen colour combinations. These results provide new insight into how individuals can learn to control attention as a function of reward contingency.     

Identifying behavioural change among drivers using Long Short-Term Memory recurrent neural networks

Globally, motor vehicle crashes account for over 1.2 million fatalities per year and are the leading cause of death for people aged 15–29 years. The majority of road crashes are caused by human error, with risk heightened among young and novice drivers learning to negotiate the complexities of the road environment. Direct feedback has been shown to have a positive impact on driving behaviour. Methods that could detect behavioural changes and therefore, positively reinforce safer driving during the early stages of driver licensing could have considerable road safety benefit. A new methodology is presented combining in-vehicle telematics technology, providing measurements forming a personalised driver profile, with neural networks to identify changes in driving behaviour. Using Long Short-Term Memory (LSTM) recurrent neural networks, individual drivers are identified based on their pattern of acceleration, deceleration and exceeding the speed limit. After model calibration, new, real-time data of the driver is supplied to the LSTM and, by monitoring prediction performance, one can assess whether a (positive or negative) change in driving behaviour is occurring over time. The paper highlights that the approach is robust to different neural network structures, data selections, calibration settings, and methodologies to select benchmarks for safe and unsafe driving. Presented case studies show additional model applications for investigating changes in driving behaviour among individuals following or during specific events (e.g., receipt of insurance renewal letters) and time periods (e.g., driving during holiday periods). The application of the presented methodology shows potential to form the basis of timely provision of direct feedback to drivers by telematics-based insurers. Such feedback may prevent internalisation of new, risky driving habits contributing to crash risk, potentially reducing deaths and injuries among young drivers as a result.     

社会交流中人际神经同步的认知机制

近年来,许多研究者开始关注社会交流中的人际神经同步机制,并将人际神经同步作为研究社会交流的一个神经指标,这对于揭示社会交流的本质和规律具有重要意义。本文从心理理论和镜像神经系统的角度,分析社会交流中神经同步的认知机制及其影响因素。未来的研究应关注这两套机制是否因交流目的、对象、形式或内容的不同,而在不同的脑区表现出神经同步,进而引发了不同认知机制的争议;以及这两套机制各自或协同工作适用的情景和任务。     

Is it worth generating rules from neural network ensembles?

Although many authors generated comprehensible models from individual networks, much less work has been done in the explanation of ensembles. DIMLP is a special neural network model from which rules are generated at the level of a single network and also at the level of an ensemble of networks. We applied ensembles of 25 DIMLP networks to several datasets of the public domain and a classification problem related to post-translational modifications of proteins. For the classification problems of the public domain, the average predictive accuracy of rulesets extracted from ensembles of neural networks was significantly better than the average predictive accuracy of rulesets generated from ensembles of decision trees. By varying the architectures of DIMLP networks we found that the average predictive accuracy of rules, as well as their complexity were quite stable. The comparison to other rule extraction techniques applied to neural networks showed that rules generated from DIMLP ensembles gave very good results. In the last problem related to bioinformatics, the best result obtained by ensembles of DIMLP networks was also significantly better than the best result obtained by ensembles of decision trees. Thus, although neural networks take much longer to train than decision trees and also rules are generated at a greater computational cost (however, still polynomial), at least for several classification problems it was worth using neural network ensembles, as extracted rules were more accurate, on average. The DIMLP software is available for PC-Linux under http://us.expasy.org/people/Guido.Bologna.html.