Learning Linear, Sparse, Factorial Codes

In previous work (Olshausen & Field 1996), an algorithm was described for learning linear sparse codes which, when trained on natural images, produces a set of basis functions that are spatially localized, oriented, and bandpass (i.e., wavelet-like). This note shows how the algorithm may be interpreted within a maximum-likelihood framework. Several useful insights emerge from this connection: it makes explicit the relation to statistical independence (i.e., factorial coding), it shows a formal relationship to the algorithm of Bell and Sejnowski (1995), and it suggests how to adapt parameters that were previously fixed.

Statistical Models for Co-occurrence Data

Modeling and predicting co-occurrences of events is a fundamental problem of unsupervised learning. In this contribution we develop a statistical framework for analyzing co-occurrence data in a general setting where elementary observations are joint occurrences of pairs of abstract objects from two finite sets. The main challenge for statistical models in this context is to overcome the inherent data sparseness and to estimate the probabilities for pairs which were rarely observed or even unobserved in a given sample set. Moreover, it is often of considerable interest to extract grouping structure or to find a hierarchical data organization. A novel family of mixture models is proposed which explain the observed data by a finite number of shared aspects or clusters. This provides a common framework for statistical inference and structure discovery and also includes several recently proposed models as special cases. Adopting the maximum likelihood principle, EM algorithms are derived to fit the model parameters. We develop improved versions of EM which largely avoid overfitting problems and overcome the inherent locality of EM--based optimization. Among the broad variety of possible applications, e.g., in information retrieval, natural language processing, data mining, and computer vision, we have chosen document retrieval, the statistical analysis of noun/adjective co-occurrence and the unsupervised segmentation of textured images to test and evaluate the proposed algorithms.

The effectiveness of assessment learning objects produced using pair programming

Pair Programming is a technique from the software development method eXtreme Programming (XP) whereby two programmers work closely together to develop a piece of software. A similar approach has been used to develop a set of Assessment Learning Objects (ALO). Three members of academic staff have developed a set of ALOs for a total of three different modules (two with overlapping content). In each case a pair programming approach was taken to the development of the ALO. In addition to demonstrating the efficiency of this approach in terms of staff time spent developing the ALOs, a statistical analysis of the outcomes for students who made use of the ALOs is used to demonstrate the effectiveness of the ALOs produced via this method.

Placing defects at the heart of high quality new homes: a learning perspective

The UK new-build housing sector is facing dual pressures to expand supply, whilst delivering against tougher planning and Building Regulation requirements; predominantly in the areas of sustainability. The sector is currently responding by significantly scaling up production and incorporating new technical solutions into new homes. This trajectory of up-scaling and technical innovation has been of research interest; but this research has primarily focus on the ‘upstream’ implications for house builders’ business models and standardised design templates. There has been little attention, though, to the potential ‘downstream’ implications of the ramping up of supply and the introduction of new technologies for build quality and defects. This paper contributes to our understanding of the ‘downstream’ implications through a synthesis of the current UK defect literature with respect to new-build housing. It is found that the prevailing emphasis in the literature is limited to the responsibility, pathology and statistical analysis of defects (and failures). The literature does not extend to how house builders individually and collectively, in practice, collect and learn from defects information. The paper concludes by describing an ongoing collaborative research programme with the National House Building Council (NHBC) to: (a) understand house builders’ localised defects analysis procedures, and their current knowledge feedback loops to inform risk management strategies; and, (b) building on this understanding, design and test action research interventions to develop new data capture, learning processes and systems to reduce targeted defects.

The impacts of learning styles, learning patterns and demographic factors on learners’ performance in a blended learning environment

Many studies have widely accepted the assumption that learning processes can be promoted when teaching styles and learning styles are well matched. In this study, the synergy between learning styles, learning patterns, and gender as a selected demographic feature and learners’ performance were quantitatively investigated in a blended learning setting. This environment adopts a traditional teaching approach of ‘one-sizefits-all’ without considering individual user’s preferences and attitudes. Hence, evidence can be provided about the value of taking such factors into account in Adaptive Educational Hypermedia Systems (AEHSs). Felder and Soloman’s Index of Learning Styles (ILS) was used to identify the learning styles of 59 undergraduate students at the University of Babylon. Five hypotheses were investigated in the experiment. Our findings show that there is no statistical significance in some of the assessed factors. However, processing dimension, the total number of hits on course website and gender indicated a statistical significance on learners’ performance. This finding needs more investigation in order to identify the effective factors on students’ achievement to be considered in Adaptive Educational Hypermedia Systems (AEHSs).

Detecting defects in UK new-build housing: a learning perspective

House builders play a key role in controlling the quality of new homes in the UK. The UK house building sector is, however, currently facing pressures to expand supply as well as conform to tougher low carbon planning and Building Regulation requirements; primarily in the areas of sustainability. There is growing evidence that the pressure the UK house building industry is currently under may be eroding build quality and causing an increase in defects. It is found that the prevailing defect literature is limited to the causes, pathology and statistical analysis of defects (and failures). The literature does not extend to examine how house builders individually and collectively, in practice, collect and learn from defects experience in order to reduce the prevalence of defects in future homes. The theoretical lens for the research is organisational learning. This paper contributes to our understanding of organisational learning in construction through a synthesis of current literature. Further, a suitable organisational learning model is adopted. The paper concludes by reporting the research design of an ongoing collaborative action research project with the National House Building Council (NHBC), focused on developing a better understanding of house builders’ localised defects analysis procedures and learning processes.

Opinion dynamics of learning agents: does seeking consensus lead to disagreement?

We study opinion dynamics in a population of interacting adaptive agents voting on a set of issues represented by vectors. We consider agents who can classify issues into one of two categories and can arrive at their opinions using an adaptive algorithm. Adaptation comes from learning and the information for the learning process comes from interacting with other neighboring agents and trying to change the internal state in order to concur with their opinions. The change in the internal state is driven by the information contained in the issue and in the opinion of the other agent. We present results in a simple yet rich context where each agent uses a Boolean perceptron to state their opinion. If the update occurs with information asynchronously exchanged among pairs of agents, then the typical case, if the number of issues is kept small, is the evolution into a society torn by the emergence of factions with extreme opposite beliefs. This occurs even when seeking consensus with agents with opposite opinions. If the number of issues is large, the dynamics becomes trapped, the society does not evolve into factions and a distribution of moderate opinions is observed. The synchronous case is technically simpler and is studied by formulating the problem in terms of differential equations that describe the evolution of order parameters that measure the consensus between pairs of agents. We show that for a large number of issues and unidirectional information flow, global consensus is a fixed point; however, the approach to this consensus is glassy for large societies.

Power-law distribution in a learning process: competition, learning and natural selection

In the present work, we propose a model for the statistical distribution of people versus number of steps acquired by them in a learning process, based on competition, learning and natural selection. We consider that learning ability is normally distributed. We found that the number of people versus step acquired by them in a learning process is given through a power law. As competition, learning and selection is also at the core of all economical and social systems, we consider that power-law scaling is a quantitative description of this process in social systems. This gives an alternative thinking in holistic properties of complex systems. (C) 2004 Elsevier B.V. All rights reserved.

Auditory processing disorder in patients with language-learning impairment and correlation with malformation of cortical development

Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)

Damage detection in a benchmark structure using AR-ARX models and statistical pattern recognition

Structural health monitoring (SHM) is related to the ability of monitoring the state and deciding the level of damage or deterioration within aerospace, civil and mechanical systems. In this sense, this paper deals with the application of a two-step auto-regressive and auto-regressive with exogenous inputs (AR-ARX) model for linear prediction of damage diagnosis in structural systems. This damage detection algorithm is based on the. monitoring of residual error as damage-sensitive indexes, obtained through vibration response measurements. In complex structures there are. many positions under observation and a large amount of data to be handed, making difficult the visualization of the signals. This paper also investigates data compression by using principal component analysis. In order to establish a threshold value, a fuzzy c-means clustering is taken to quantify the damage-sensitive index in an unsupervised learning mode. Tests are made in a benchmark problem, as proposed by IASC-ASCE with different damage patterns. The diagnosis that was obtained showed high correlation with the actual integrity state of the structure. Copyright © 2007 by ABCM.

A statistical analysis of loss factor to determine the energy losses

Given that the total amount of losses in a distribution system is known, with a reliable methodology for the technical loss calculation, the non-technical losses can be obtained by subtraction. A usual method of calculation technical losses in the electric utilities uses two important factors: load factor and the loss factor. The load factor is usually obtained with energy and demand measurements, whereas, to compute the loss factor it is necessary the learning of demand and energy loss, which are not, in general, prone of direct measurements. In this work, a statistical analysis of this relationship using the curves of a sampling of consumers in a specific company is presented. These curves will be summarized in different bands of coefficient k. Then, it will be possible determine where each group of consumer has its major concentration of points. ©2008 IEEE.

Power Electronics course: Analysis and evaluation of the educational software and the environment learning

This paper presents the analysis and evaluation of the Power Electronics course at So Paulo State University-UNESP-Campus of Ilha Solteira(SP)-Brazil, which includes the usage of interactive Java simulations tools and an educational software to aid the teaching of power electronic converters. This platform serves as an oriented course for the lectures and supplementary support for laboratory experiments in the power electronics courses. The simulation tools provide an interactive and dynamic way to visualize the power electronics converters behavior together with the educational software, which contemplates the theory and a list of subjects for circuit simulations. In order to verify the performance and the effectiveness of the proposed interactive educational platform, it is presented a statistical analysis considering the last three years. © 2011 IEEE.

Mensuração de estratégias de estudo e aprendizagem de alunos universitários: learning and study strategies inventory (lassi) adaptação e validação para o Brasil

Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)

Effects of a metatextual intervention program in developing written narratives of students with learning difficulties

This study aimed to verify the effects of a metatextual intervention program, in the elaboration of stories written by students with learning difficulties. Four students were included in the sample of both genders, with ages ranging between eight years and four months and ten years and two months of age. The program was implemented at the participant schools, using an approach of multiple baseline within-subjects, with two conditions: baseline and intervention. Data analysis was based on the classification of stories produced by the students. Mann-Whitney testing was also applied, to analyze whether there have been significant changes in these productions. The results indicated that all students have improved performance in relation to the categories of produced stories, from elementary schemas (33%), for a more elaborate scheme (77%), with a better structuring of the elements that constitute a story. Statistical analysis also showed that the intervention has produced significant results for all variables analyzed. The data obtained have shown that the program was effective.

Effect of short training on vaginal fluid microscopy (wet mount) learning

Objective: Is it feasible to learn the basics of wet mount microscopy of vaginal fluid in 10 hours?Materials and Methods: This is a pilot project wherein 6 students with different grades of education were invited for being tested on their ability to read wet mount microscopic slides before and after 10 hours of hands-on training. Microscopy was performed according to a standard protocol (Femicare, Tienen, Belgium). Before and after training, all students had to evaluate a different set of 50 digital slides. Different diagnoses and microscopic patterns had to be scored. kappa indices were calculated compared with the expert reading. Results: All readers improved their mean scores significantly, especially for the most important types of altered flora (p < .0001). The mean increase in reading concordance (kappa from 0.64 to 0.75) of 1 student with a solid previous experience with microscopy did not reach statistical significance, but the remaining 5 students all improved their scores from poor performance (all kappa < 0.20) to moderate (kappa = 0.53, n = 1) to good (kappa > 0.61, n = 4) concordance. Reading quality improved and reached fair to good concordance on all microscopic items studied, except for the detection of parabasal cells and cytolytic flora. Conclusions: Although further improvement is still possible, a short training course of 10 hours enables vast improvement on wet mount microscopy accuracy and results in fair to good concordance of the most important variables of the vaginal flora compared to a reference reader.