Connectionism and the learning of probabilistic concepts

Three experiments examine the claim of Gluck and Bower (1986, 1988a, 1988b) that the learning of medical concepts can be simulated by a connectionist network in which the symptoms are the input and the diagnosis is the output. The first experiment replicates the main finding of Gluck and Bower. In this experiment, subjects were required to estimate the probability of each of two diseases, given a particular target symptom. In fact these two probabilities were identical, but because one illness was more common than the other, the target symptom was a better predictor of the rare disease than of the common disease. Contrary to a normative probability judgement account, subjects were biased in that they judged the probability of the rare disease given the target symptom to be greater than the probability of the common disease given the target symptom. Gluck and Bower argued that such a result was predicted by a connectionist network using the Rescorla-Wagner learning rule, but it is argued that Gluck and Bower's network simulation was not appropriate for the experiment they had performed. In fact, it appeared that the connectionist network failed to predict the bias in the subjects' probability estimates.

However, this conclusion rests on an assumption that Gluck and Bower implicitly made. They arranged for P(rare disease/target symptom) and P(com-mon disease/target symptom) to be identical across all trials on which the target symptom occurred, both on its own and with other symptoms present. Gluck and Bower assumed that the subjects were estimating these probabilities. But the results of the second experiment showed instead that the subjects were estimating the probability of each disease given only the target symptom. In the final experiment the design was changed so that this problem might be circumvented. In this experiment, again, the subjects were biased in their probability judgements exactly as the connectionist network predicted. Thus, finally, evidence was found which was compatible with the network model but not with a normative account, but this was true only if the network did not include a layer of hidden units.     

Use of evidence in a categorization task: analytic and holistic processing modes

Category learning performance can be influenced by many contextual factors, but the effects of these factors are not the same for all learners. The present study suggests that these differences can be due to the different ways evidence is used, according to two main basic modalities of processing information, analytically or holistically. In order to test the impact of the information provided, an inductive rule-based task was designed, in which feature salience and comparison informativeness between examples of two categories were manipulated during the learning phases, by introducing and progressively reducing some perceptual biases. To gather data on processing modalities, we devised the Active Feature Composition task, a production task that does not require classifying new items but reproducing them by combining features. At the end, an explicit rating task was performed, which entailed assessing the accuracy of a set of possible categorization rules. A combined analysis of the data collected with these two different tests enabled profiling participants in regard to the kind of processing modality, the structure of representations and the quality of categorial judgments. Results showed that despite the fact that the information provided was the same for all participants, those who adopted analytic processing better exploited evidence and performed more accurately, whereas with holistic processing categorization is perfectly possible but inaccurate. Finally, the cognitive implications of the proposed procedure, with regard to involved processes and representations, are discussed.     

Inference processes in learning well-defined concepts

In an adaptation of usual experimental techniques for studying the performance of a subject who has to discover a well-defined concept, two theories were independently tested. It was found that Bourne's set of inference operations (Bourne 1974; Salatas and Bourne 1974) predicted fairly well the relative difficulty for the connectives studied (inclusive disjunctive, alternative denial, joint denial, exclusive, exclusive disjunctive) for a stimulus population based on three-valued dimensions but not for a four-valued stimulus population. Reaction times gathered after the subject made no further classification errors fitted the decision-tree structures developed by Hunt et al. (1966). Inspection of the hypotheses subjects made about the rule, revealed two possible major reasoning obstacles: the introduction of a negative attribute and the notion of exception i.e. when the assignment of an attribute in one class depends on another attribute whereas in another class it does not.     

When category learning is holistic: A reply to Ward and Scott

A reply is presented to Ward and Scott’s (1987) recent reservations about the evidence for Kemler Nelson’s (1984) claims about when category learning is likely to be holistic. Focusing on the effect of intention, this paper suggests that: (1) contrary to Ward and Scott’s contention, a reanalysis of a critical set of original data continues to support Kemler Nelson’s claim of more holistic learning under unintentional conditions; (2) there is converging evidence for that claim; (3) Ward and Scott’s incidental learning data may diverge because of the inclusion of many weak learners; (4) Ward and Scott’s counterproposal makes some implausible and unsupported predictions; and (5) some of Ward and Scott’s reaction-time data are difficult to interpret. Still, a final discussion identifies some significant points of agreement with Ward and Scott.     

Two modes of transfer in artificial grammar learning

Participants can transfer grammatical knowledge acquired implicitly in 1 vocabulary to new sequences instantiated in both the same and a novel vocabulary. Two principal theories have been advanced to account for these effects. One suggests that sequential dependencies form the basis for cross-domain transfer (e.g., Z. Dienes, G. T. M. Altmann, & S. J. Gao, 1999). Another argues that a form of episodic memory known as abstract analogy is sufficient (e.g., L. R. Brooks & J. R. Vokey, 1991). Three experiments reveal the contributions of the 2. In Experiment 1 sequential dependencies form the only basis for transfer. Experiment 2 demonstrates that this process is impaired by a change in the distributional properties of the language. Experiment 3 demonstrates that abstract analogy of repetition structure is relatively immune to such a change. These findings inform theories of artificial grammar learning and the transfer of grammatical knowledge.     

Evidence for the effectiveness of holistic process goals for learning and performance under pressure

ObjectivesResearch has suggested that holistic process goals might help avoid the effects associated with conscious processing of task relevant information by skilled but anxious athletes. This experiment compared the efficacy of holistic and part process goal strategies for novices using a learning paradigm.DesignLaboratory-based experimental design incorporating practice, retention and transfer phases.MethodTwenty-four males were randomly assigned to a part process goal, holistic process goal or control condition and performed a simulated race-driving task in practice, retention and transfer tests.ResultsAnalyses of variance revealed that performance during practice was similar in all conditions but that the holistic process goal group outperformed the part process goal group at both retention and transfer.ConclusionsCompared to part process goals, holistic process goals result in more effective motor learning and performance that appears to be more robust under pressure.     

Assessing the effect of cognitive styles with different learning modes on learning outcome

In this study, similarities and differences in learning outcome associated with individual differences in cognitive styles are examined using the traditional (face-to-face) and web-based learning modes. 140 undergraduate students were categorized as having analytic or holistic cognitive styles by their scores on the Style of Learning and Thinking questionnaire. Four different conditions were studies; students with analytic cognitive style in a traditional learning mode, analytic cognitive style in a web-based learning mode, holistic cognitive style in a traditional learning mode, and holistic cognitive style in a web-based learning mode. Analysis of the data show that analytic style in traditional mode lead to significantly higher performance and perceived satisfaction than in other conditions. Satisfaction did not differ significantly between students with analytic style in web-based learning and those with holistic style in traditional learning. This suggest that integrating different learning modes into the learning environment may be insufficient to improve learners' satisfaction.     

易学思维之葩掇拾

本文选取《易》四卦探析其思维特征:从《泰》、《否》剖辨其因果思维,《泰》之以"果"示"因"更属辩证逻辑思维,从而否定C·G荣格关于《易》不从"因果关系论事"的判断。以《蛊》卦之"蛊"不训"弊乱"而应训如钱钟书先生所证之"事"、"事业",论定《蛊》属于创造性思维范畴,继承前辈事业,在自己的努力实践中创新业,立新功。以《》卦辞"不利即戎"断句或连读,探析前人关于战争的两种思考,一是果决地抓住战机打击敌人,如周武王讨伐纣之《牧誓》所示;一是先安顿内部,"施禄及下"以得民心,如《左传》"曹刿论战"所示,是"攘外必先安内"的战略性思考。     

Analytic Sadism and Analytic Restraint

Csillag challenges us to consider the presence and impact of the analyst’s sadism in the clinical encounter. Her theme usefully adds to the literature on the analyst’s countertransference and pushes us to look harder at ourselves. I distinguish between sadistic intent and experienced sadistic impact and suggest that Csillag is mostly speaking about the latter. Additionally, I address the absence of analytic restraint as a professional ideal that might moderate our slide to sadism in the countertransference.     

Re-framing the characteristics of concepts and their relation to learning and cognition in artificial agents

In this work, the problems of knowledge acquisition and information processing are explored in relation to the definitions of concepts and conceptual processing, and their implications for artificial agents.The discussion focuses on views of cognition as a dynamic property in which the world is actively represented in grounded mental states which only have meaning in the action context. Reasoning is understood as an emerging property consequence of actions-environment couplings achieved through experience, and concepts as situated and dynamic phenomena enabling behaviours.Re-framing the characteristics of concepts is considered crucial to overcoming settled beliefs and reinterpreting new understandings in artificial systems.The first part presents a review of concepts from cognitive sciences. Support is found for views on grounded and embodied cognition, describing concepts as dynamic, flexible, context-dependent, and distributedly coded.That is argued to contrast with many technical implementations assuming concepts as categories, whilst explains limitations when grounding amodal symbols, or in unifying learning, perception and reasoning.The characteristics of concepts are linked to methods of active inference, self-organization, and deep learning to address challenges posed and to reinterpret emerging techniques.In a second part, an architecture based on deep generative models is presented to illustrate arguments elaborated. It is evaluated in a navigation task, showing that sufficient representations are created regarding situated behaviours with no semantics imposed on data. Moreover, adequate behaviours are achieved through a dynamic integration of perception and action in a single representational domain and process.