The evolution of learning systems:to Bayes or not to be

Bayesian algorithms pose a limit to the performance learning algorithms can achieve. Natural selection should guide the evolution of information processing systems towards those limits. What can we learn from this evolution and what properties do the intermediate stages have? While this question is too general to permit any answer, progress can be made by restricting the class of information processing systems under study. We present analytical and numerical results for the evolution of on-line algorithms for learning from examples for neural network classifiers, which might include or not a hidden layer. The analytical results are obtained by solving a variational problem to determine the learning algorithm that leads to maximum generalization ability. Simulations using evolutionary programming, for programs that implement learning algorithms, confirm and expand the results. The principal result is not just that the evolution is towards a Bayesian limit. Indeed it is essentially reached. In addition we find that evolution is driven by the discovery of useful structures or combinations of variables and operators. In different runs the temporal order of the discovery of such combinations is unique. The main result is that combinations that signal the surprise brought by an example arise always before combinations that serve to gauge the performance of the learning algorithm. This latter structures can be used to implement annealing schedules. The temporal ordering can be understood analytically as well by doing the functional optimization in restricted functional spaces. We also show that there is data suggesting that the appearance of these traits also follows the same temporal ordering in biological systems. © 2006 American Institute of Physics.

A theoretical and experimental account of n-tuple classifier performance

The n-tuple recognition method is briefly reviewed, summarizing the main theoretical results. Large-scale experiments carried out on Stat-Log project datasets confirm this method as a viable competitor to more popular methods due to its speed, simplicity, and accuracy on the majority of a wide variety of classification problems. A further investigation into the failure of the method on certain datasets finds the problem to be largely due to a mismatch between the scales which describe generalization and data sparseness.

Differential impact of posterior lesions in the left and right hemisphere on visual category learning and generalization to contrast reversal

Hemispheric differences in the learning and generalization of pattern categories were explored in two experiments involving sixteen patients with unilateral posterior, cerebral lesions in the left (LH) or right (RH) hemisphere. In each experiment participants were first trained to criterion in a supervised learning paradigm to categorize a set of patterns that either consisted of simple geometric forms (Experiment 1) or unfamiliar grey-level images (Experiment 2). They were then tested for their ability to generalize acquired categorical knowledge to contrast-reversed versions of the learning patterns. The results showed that RH lesions impeded category learning of unfamiliar grey-level images more severely than LH lesions, whereas this relationship appeared reversed for categories defined by simple geometric forms. With regard to generalization to contrast reversal, categorization performance of LH and RH patients was unaffected in the case of simple geometric forms. However, generalization to of contrast-reversed grey-level images distinctly deteriorated for patients with LH lesions relative to those with RH lesions, with the latter (but not the former) being consistently unable to identify the pattern manipulation. These findings suggest a differential use of contrast information in the representation of pattern categories in the two hemispheres. Such specialization appears in line with previous distinctions between a predominantly lefthemispheric, abstract-analytical and a righthemispheric, specific-holistic representation of object categories, and their prediction of a mandatory representation of contrast polarity in the RH. Some implications for the well-established dissociation of visual disorders for the recognition of faces and letters are discussed.

The impact of learning outcomes in business education: assessing value, relevance and graduate ability in a multi-country study of employers and business graduates

This paper discusses critical findings from a two-year EU-funded research project involving four European countries: Austria, England, Slovenia and Romania. The project had two primary aims. The first of these was to develop a systematic procedure for assessing the balance between learning outcomes acquired in education and the specific needs of the labour market. The second aim was to develop and test a set of meta-level quality indicators aimed at evaluating the linkages between education and employment. The project was distinctive in that it combined different partners from Higher Education, Vocational Training, Industry and Quality Assurance. One of the key emergent themes identified in exploratory interviews was that employers and recent business graduates in all four countries want a well-rounded education which delivers a broad foundation of key business knowledge across the various disciplines. Both groups also identified the need for personal development in critical skills and competencies. Following the exploratory study, a questionnaire was designed to address five functional business areas, as well as a cluster of 8 business competencies. Within the survey, questions relating to the meta-level quality indicators assessed the impact of these learning outcomes on the workplace, in terms of the following: 1) value, 2) relevance and 3) graduate ability. This paper provides an overview of the study findings from a sample of 900 business graduates and employers. Two theoretical models are proposed as tools for predicting satisfaction with work performance and satisfaction with business education. The implications of the study findings for education, employment and European public policy are discussed.

The n-tuple classifier: too good to ignore

The n-tuple pattern recognition method has been tested using a selection of 11 large data sets from the European Community StatLog project, so that the results could be compared with those reported for the 23 other algorithms the project tested. The results indicate that this ultra-fast memory-based method is a viable competitor with the others, which include optimisation-based neural network algorithms, even though the theory of memory-based neural computing is less highly developed in terms of statistical theory.

Benchmarking the n-tuple classifier with statlog dataset

The n-tuple recognition method was tested on 11 large real-world data sets and its performance compared to 23 other classification algorithms. On 7 of these, the results show no systematic performance gap between the n-tuple method and the others. Evidence was found to support a possible explanation for why the n-tuple method yields poor results for certain datasets. Preliminary empirical results of a study of the confidence interval (the difference between the two highest scores) are also reported. These suggest a counter-intuitive correlation between the confidence interval distribution and the overall classification performance of the system.

Good-turing estimation for the frequentist n-tuple classifier

We present results concerning the application of the Good-Turing (GT) estimation method to the frequentist n-tuple system. We show that the Good-Turing method can, to a certain extent rectify the Zero Frequency Problem by providing, within a formal framework, improved estimates of small tallies. We also show that it leads to better tuple system performance than Maximum Likelihood estimation (MLE). However, preliminary experimental results suggest that replacing zero tallies with an arbitrary constant close to zero before MLE yields better performance than that of GT system.

Noise, regularizers, and unrealizable scenarios in online learning from restricted training sets

We study the dynamics of on-line learning in multilayer neural networks where training examples are sampled with repetition and where the number of examples scales with the number of network weights. The analysis is carried out using the dynamical replica method aimed at obtaining a closed set of coupled equations for a set of macroscopic variables from which both training and generalization errors can be calculated. We focus on scenarios whereby training examples are corrupted by additive Gaussian output noise and regularizers are introduced to improve the network performance. The dependence of the dynamics on the noise level, with and without regularizers, is examined, as well as that of the asymptotic values obtained for both training and generalization errors. We also demonstrate the ability of the method to approximate the learning dynamics in structurally unrealizable scenarios. The theoretical results show good agreement with those obtained by computer simulations.

Evaluation of an open-field autorefractor's ability to measure refraction and hence potential to assess objective accommodation in pseudophakes

Background: To evaluate the accuracy of an open-field autorefractor compared with subjective refraction in pseudophakes and hence its ability to assess objective eye focus with intraocular lenses (IOLs). Methods: Objective refraction was measured at 6 m using the Shin-Nippon NVision-K 5001/Grand Seiko WR-5100K open-field autorefractor (five repeats) and by subjective refraction on 141 eyes implanted with a spherical (Softec1 n=53), aspherical (SoftecHD n=37) or accommodating (1CU n=22; Tetraflex n=29) IOL. Autorefraction was repeated 2 months later. Results: The autorefractor prescription was similar (average difference: 0.09±0.53 D; p=0.19) to that found by subjective refraction, with ~71% within ±0.50 D. The horizontal cylindrical components were similar (difference: 0.00±0.39 D; p=0.96), although the oblique (J45) autorefractor cylindrical vector was slightly more negative (by -0.06±0.25 D; p=0.06) than the subjective refraction. The results were similar for each of the IOL designs except for the spherical IOL, where the mean spherical equivalent difference between autorefraction and subjective was more hypermetropic than the Tetraflex accommodating IOL (F=2.77, p=0.04). The intrasession repeatability was

Are faces special? A case of pure prosopagnosia

The ability to recognize individual faces is of crucial social importance for humans and evolutionarily necessary for survival. Consequently, faces may be “special” stimuli, for which we have developed unique modular perceptual and recognition processes. Some of the strongest evidence for face processing being modular comes from cases of prosopagnosia, where patients are unable to recognize faces whilst retaining the ability to recognize other objects. Here we present the case of an acquired prosopagnosic whose poor recognition was linked to a perceptual impairment in face processing. Despite this, she had intact object recognition, even at a subordinate level. She also showed a normal ability to learn and to generalize learning of nonfacial exemplars differing in the nature and arrangement of their parts, along with impaired learning and generalization of facial exemplars. The case provides evidence for modular perceptual processes for faces.

Challenging the predictive ability of accounting techniques in modelling organizational futures

This article considers the role of accounting in organisational decision making. It challenges the rational nature of decisions made in organisations through the use of accounting models and the problems of predicting the future through the use of such models. The use of accounting in this manner is evaluated from an epochal postmodern stance. Issues raised by chaos theory and the uncertainty principle are used to demonstrate problems with the predictive ability of accounting models. The authors argue that any consideration of the predictive value of accounting needs to change to incorporate a recognition of the turbulent external environment, if it is to be of use for organisational decision making. Thus it is argued that the role of accounting as a mechanism for knowledge creation regarding the future is fundamentally flawed. We take this as a starting-point to argue for the real purpose of the use of the predictive techniques of accounting, using its ritualistic role in the context of myth creation to argue for the cultural benefits of the use of such flawed techniques.

Applications of the behavioural homogeneity evaluation framework: the predictive ability of consumer involvement for international food market segmentation

This article proposes a framework of alternative international marketing strategies, based on the evaluation of intra- and inter-cultural behavioural homogeneity for market segmentation. The framework developed in this study provides a generic structure to behavioural homogeneity, proposing consumer involvement as a construct with unique predictive ability for international marketing strategy decisions. A model-based segmentation process, using structural equation models, is implemented to illustrate the application of the framework.

Ability of dyslexic and control teenagers to sustain attention and inhibit responses

Dyslexia and attentional difficulty have often been linked, but little is known of the nature of the supposed attentional disorder. The Sustained Attention to Response Task (SART: Robertson, Manly, Andrade, Baddeley and Yiend, 1997) was designed as a measure of sustained attention and requires the withholding of responses to rare (one in nine) targets. To investigate the nature of the attentional disorder in dyslexia, this paper reports two studies which examined the performance of teenagers with dyslexia and their age-matched controls on the SART, the squiggle SART (a modification of the SART using novel and unlabellable stimuli rather than digits) and the go-gap-stop test of response inhibition (GGST). Teenagers with dyslexia made significantly more errors than controls on the original SART, but not the squiggle SART. There were no group differences on the GGST. After controlling for speed of reaction time in a sequential multiple regression predicting SART false alarms, false alarms on the GGST accounted for up to 22% extra variance in the control groups (although less on the squiggle SART) but negligible amounts of variance in the dyslexic groups. We interpret the results as reflecting a stimulus recognition automaticity deficit in dyslexia, rather than a sustained attention deficit. Furthermore, results suggest that response inhibition is an important component of performance on the standard SART when stimuli are recognised automatically.