Deep learning of representations and its application to computer vision

L’objectif de cette thèse par articles est de présenter modestement quelques étapes du parcours qui mènera (on espère) à une solution générale du problème de l’intelligence artificielle. Cette thèse contient quatre articles qui présentent chacun une différente nouvelle méthode d’inférence perceptive en utilisant l’apprentissage machine et, plus particulièrement, les réseaux neuronaux profonds. Chacun de ces documents met en évidence l’utilité de sa méthode proposée dans le cadre d’une tâche de vision par ordinateur. Ces méthodes sont applicables dans un contexte plus général, et dans certains cas elles on tété appliquées ailleurs, mais ceci ne sera pas abordé dans le contexte de cette de thèse. Dans le premier article, nous présentons deux nouveaux algorithmes d’inférence variationelle pour le modèle génératif d’images appelé codage parcimonieux “spike- and-slab” (CPSS). Ces méthodes d’inférence plus rapides nous permettent d’utiliser des modèles CPSS de tailles beaucoup plus grandes qu’auparavant. Nous démontrons qu’elles sont meilleures pour extraire des détecteur de caractéristiques quand très peu d’exemples étiquetés sont disponibles pour l’entraînement. Partant d’un modèle CPSS, nous construisons ensuite une architecture profonde, la machine de Boltzmann profonde partiellement dirigée (MBP-PD). Ce modèle a été conçu de manière à simplifier d’entraînement des machines de Boltzmann profondes qui nécessitent normalement une phase de pré-entraînement glouton pour chaque couche. Ce problème est réglé dans une certaine mesure, mais le coût d’inférence dans le nouveau modèle est relativement trop élevé pour permettre de l’utiliser de manière pratique. Dans le deuxième article, nous revenons au problème d’entraînement joint de machines de Boltzmann profondes. Cette fois, au lieu de changer de famille de modèles, nous introduisons un nouveau critère d’entraînement qui donne naissance aux machines de Boltzmann profondes à multiples prédictions (MBP-MP). Les MBP-MP sont entraînables en une seule étape et ont un meilleur taux de succès en classification que les MBP classiques. Elles s’entraînent aussi avec des méthodes variationelles standard au lieu de nécessiter un classificateur discriminant pour obtenir un bon taux de succès en classification. Par contre, un des inconvénients de tels modèles est leur incapacité de générer deséchantillons, mais ceci n’est pas trop grave puisque la performance de classification des machines de Boltzmann profondes n’est plus une priorité étant donné les dernières avancées en apprentissage supervisé. Malgré cela, les MBP-MP demeurent intéressantes parce qu’elles sont capable d’accomplir certaines tâches que des modèles purement supervisés ne peuvent pas faire, telles que celle de classifier des données incomplètes ou encore celle de combler intelligemment l’information manquante dans ces données incomplètes. Le travail présenté dans cette thèse s’est déroulé au milieu d’une période de transformations importantes du domaine de l’apprentissage à réseaux neuronaux profonds qui a été déclenchée par la découverte de l’algorithme de “dropout” par Geoffrey Hinton. Dropout rend possible un entraînement purement supervisé d’architectures de propagation unidirectionnel sans être exposé au danger de sur- entraînement. Le troisième article présenté dans cette thèse introduit une nouvelle fonction d’activation spécialement con ̧cue pour aller avec l’algorithme de Dropout. Cette fonction d’activation, appelée maxout, permet l’utilisation de aggrégation multi-canal dans un contexte d’apprentissage purement supervisé. Nous démontrons comment plusieurs tâches de reconnaissance d’objets sont mieux accomplies par l’utilisation de maxout. Pour terminer, sont présentons un vrai cas d’utilisation dans l’industrie pour la transcription d’adresses de maisons à plusieurs chiffres. En combinant maxout avec une nouvelle sorte de couche de sortie pour des réseaux neuronaux de convolution, nous démontrons qu’il est possible d’atteindre un taux de succès comparable à celui des humains sur un ensemble de données coriace constitué de photos prises par les voitures de Google. Ce système a été déployé avec succès chez Google pour lire environ cent million d’adresses de maisons.

Continuous improvement programs as processes for organizational learning: a case study of a company in the food products industry

This article intends to show the relationships between quality practices and the process of organizational learning. When we look at the literature about programs of continuous improvement we see that theoreticians consider that the process of organizational learning is a superior stage in the quality culture adopted by companies. To investigate this possibility, we put together a series of indicators taken from classic authors who have written about organizational learning. Adopting a multiple methodology, we applied these indicators to two plants belonging to the Nestlé food product company which have introduced continuous improvement programs over the last two years.

The role of context in forming young learners' attitudes and motivation to learning French

Despite the wealth of valuable information that has been generated by motivation studies to date, there are certain limitations in the common approaches. Quantitative and psychometric approaches to motivation research that have dominated in recent decades provided epiphenomenal descriptions of learner motivation within different contexts. However, these approaches assume homogeneity within a given group and often mask the variation between learners within the same, and different, contexts. Although these studies have provided empirical data to form and validate theoretical constructs, they have failed to recognise learners as individual ‘people’ that interact with their context. Learning context has become increasingly explicit in motivation studies, (see Coleman et al. 2007 and Housen et al. 2011), however it is generally considered as a background variable which is pre-existing and external to the individual. Stemming from the recent ‘social turn’ (Block 2003) in SLA research from a more cognitive-linguistic perspective to a more context-specific view of language learning, there has been an upsurge in demand for a greater focus on the ‘person in context’ in motivation research (Ushioda 2011). This paper reports on the findings of a longitudinal study of young English learners of French as they transition from primary to secondary school. Over 12 months, the study employed a mixed-method approach in order to gain an in-depth understanding of how the learners’ context influenced attitudes to language learning. The questionnaire results show that whilst the learners displayed some consistent and stable motivational traits over the 12 months, there were significant differences for learners within different contexts in terms of their attitudes to the language classroom and their levels of self-confidence. A subsequent examination of the qualitative focus group data provided an insight into how and why these attitudes were formed and emphasised the dynamic and complex interplay between learners and their context.

State of the art of learning styles-based adaptive educational hypermedia systems (Ls-Baehss)

The notion that learning can be enhanced when a teaching approach matches a learner’s learning style has been widely accepted in classroom settings since the latter represents a predictor of student’s attitude and preferences. As such, the traditional approach of ‘one-size-fits-all’ as may be applied to teaching delivery in Educational Hypermedia Systems (EHSs) has to be changed with an approach that responds to users’ needs by exploiting their individual differences. However, establishing and implementing reliable approaches for matching the teaching delivery and modalities to learning styles still represents an innovation challenge which has to be tackled. In this paper, seventy six studies are objectively analysed for several goals. In order to reveal the value of integrating learning styles in EHSs, different perspectives in this context are discussed. Identifying the most effective learning style models as incorporated within AEHSs. Investigating the effectiveness of different approaches for modelling students’ individual learning traits is another goal of this study. Thus, the paper highlights a number of theoretical and technical issues of LS-BAEHSs to serve as a comprehensive guidance for researchers who interest in this area.

Learning of shared attention in sociable robotics

Sociable robots are embodied agents that are part of a heterogeneous society of robots and humans. They Should be able to recognize human beings and each other, and to engage in social, interactions. The use of a robotic architecture may strongly reduce the time and effort required to construct a sociable robot. Such architecture must have structures and mechanisms to allow social interaction. behavior control and learning from environment. Learning processes described oil Science of Behavior Analysis may lead to the development of promising methods and Structures for constructing robots able to behave socially and learn through interactions from the environment by a process of contingency learning. In this paper, we present a robotic architecture inspired from Behavior Analysis. Methods and structures of the proposed architecture, including a hybrid knowledge representation. are presented and discussed. The architecture has been evaluated in the context of a nontrivial real problem: the learning of the shared attention, employing an interactive robotic head. The learning capabilities of this architecture have been analyzed by observing the robot interacting with the human and the environment. The obtained results show that the robotic architecture is able to produce appropriate behavior and to learn from social interaction. (C) 2009 Elsevier Inc. All rights reserved.

Reading, equivalence, and recombination of units: A replication with students with different learning histories

First graders, preschoolers, special education students, and adults received a reading program in which they learned to match printed to dictated words and to construct (copy) printed words. The students not only learned to match the training words but also learned to read them. In addition, most of the students learned to read new words that involved recombinations of the syllables of the training words. The results replicate and extend the generality of a prior analysis of a reading program based on stimulus equivalence and recombination of units.

Learning Curves of Temporary Workers in Manual Order Picking Activities

Person-to-stock order picking is highly flexible and requires minimal investment costs in comparison to automated picking solutions. For these reasons, tradi-tional picking is widespread in distribution and production logistics. Due to its typically large proportion of manual activities, picking causes the highest operative personnel costs of all intralogistics process. The required personnel capacity in picking varies short- and mid-term due to capacity requirement fluctuations. These dynamics are often balanced by employing minimal permanent staff and using seasonal help when needed. The resulting high personnel fluctuation necessitates the frequent training of new pickers, which, in combination with in-creasingly complex work contents, highlights the im-portance of learning processes in picking. In industrial settings, learning is often quantified based on diminishing processing time and cost requirements with increasing experience. The best-known industrial learning curve models include those from Wright, de Jong, Baloff and Crossman, which are typically applied to the learning effects of an entire work crew rather than of individuals. These models have been validated in largely static work environments with homogeneous work contents. Little is known of learning effects in picking systems. Here, work contents are heterogeneous and individual work strategies vary among employees. A mix of temporary and steady employees with varying degrees of experience necessitates the observation of individual learning curves. In this paper, the individual picking performance development of temporary employees is analyzed and compared to that of steady employees in the same working environment.

Context Aware e-Learning Objects and their Types from the Perspective of the Object-Oriented Paradigm

The contribution of this article demonstrates how to identify context-aware types of e-Learning objects (eLOs) derived from the subject domains. This perspective is taken from an engineering point of view and is applied during requirements elicitation and analysis relating to present work in constructing an object-oriented (OO), dynamic, and adaptive model to build and deliver packaged e-Learning courses. Consequently, three preliminary subject domains are presented and, as a result, three primitive types of eLOs are posited. These types educed from the subject domains are of structural, conceptual, and granular nature. Structural objects are responsible for the course itself, conceptual objects incorporate adaptive and logical interoperability, while granular objects congregate granular assets. Their differences, interrelationships, and responsibilities are discussed. A major design challenge relates to adaptive behaviour. Future research addresses refinement on the subject domains and adaptive hypermedia systems.

The Comprehensive Assessment of Leadership for Learning: A Next Generation Formative Evaluation and Feedback System

Recent developments in federal policy have prompted the creation of state evaluation frameworks for principals and teachers that hold educators accountable for effective practices and student outcomes. These changes have created a demand for formative evaluation instruments that reflect current accountability pressures and can be used by schools to focus school improvement and leadership development efforts. The Comprehensive Assessment of Leadership for Learning (CALL) is a next generation, 360-degree on-line assessment and feedback system that reflect best practices in feedback design. Some unique characteristics of CALL include a focus on: leadership distributed throughout the school rather than as carried out by an individual leader; assessment of leadership tasks rather than perceptions of leadership practice; a focus on larger complex systems of middle and high school; and transparency of assessment design. This paper describes research contributing to the design and validation of the CALL survey instrument.

Expanding Learning Time in Schools: Considering the Challenges of Implementation and the Potential Impact

Considering the broader context of school reform that is seeking education strategies that might deliver substantial impact, this article examines four questions related to the policy and practice of expanding learning time: (a) why do educators find the standard American school calendar insufficient to meet students’ educational needs, especially those of disadvantaged students? (b) how do educators implement a longer day and/or year, addressing concerns about both educational quality and costs? (c) what does research report about outcomes of expanding time in schools? and (d) what are the future prospects for increasing the number of expanded-time schools? The paper examines these questions by considering research, policy, and practice at the national level and, throughout, by drawing upon additional evidence from Massachusetts, one of the leading states in the expanded-time movement. In considering the latter two questions, the article explores the knowns and unknowns related to expanded learning time and offers suggestions for further research.

A Step in the Right Direction: Development of the Computerized Assessment for Preschool Social Emotional Learning (CAPSEL)

Invited commentary on "Computerizing Social-Emotional Assessment for School Readiness".

The Learning of the Mathematical Models to Aid Decision Making in the High Level Engineering Schools?

At present, in the University curricula in most countries, the decision theory and the mathematical models to aid decision making is not included, as in the graduate program like in Doctored and Master´s programs. In the Technical School of High Level Agronomic Engineers of the Technical University of Madrid (ETSIA-UPM), the need to offer to the future engineers training in a subject that could help them to take decisions in their profession was felt. Along the life, they will have to take a lot of decisions. Ones, will be important and others no. In the personal level, they will have to take several very important decisions, like the election of a career, professional work, or a couple, but in the professional field, the decision making is the main role of the Managers, Politicians and Leaders. They should be decision makers and will be paid for it. Therefore, nobody can understand that such a professional that is called to practice management responsibilities in the companies, does not take training in such an important matter. For it, in the year 2000, it was requested to the University Board to introduce in the curricula an optional qualified subject of the second cycle with 4,5 credits titled " Mathematical Methods for Making Decisions ". A program was elaborated, the didactic material prepared and programs as Maple, Lingo, Math Cad, etc. installed in several IT classrooms, where the course will be taught. In the course 2000-2001 this subject was offered with a great acceptance that exceeded the forecasts of capacity and had to be prepared more classrooms. This course in graduate program took place in the Department of Applied Mathematics to the Agronomic Engineering, as an extension of the credits dedicated to Mathematics in the career of Engineering.

Knowledge spirals, situational leadership and informal learning applied on b-learning methodology

The methodology “b-learning” is a new teaching scenario and it requires the creation, adaptation and application of new learning tools searching the assimilation of new collaborative competences. In this context, it is well known the knowledge spirals, the situational leadership and the informal learning. The knowledge spirals is a basic concept of the knowledge procedure and they are based on that the knowledge increases when a cycle of 4 phases is repeated successively.1) The knowledge is created (for instance, to have an idea); 2) The knowledge is decoded into a format to be easily transmitted; 3) The knowledge is modified to be easily comprehensive and it is used; 4) New knowledge is created. This new knowledge improves the previous one (step 1). Each cycle shows a step of a spiral staircase: by going up the staircase, more knowledge is created. On the other hand, the situational leadership is based on that each person has a maturity degree to develop a specific task and this maturity increases with the experience. Therefore, the teacher (leader) has to adapt the teaching style to the student (subordinate) requirements and in this way, the professional and personal development of the student will increase quickly by improving the results and satisfaction. This educational strategy, finally combined with the informal learning, and in particular the zone of proximal development, and using a learning content management system own in our University, gets a successful and well-evaluated learning activity in Master subjects focused on the collaborative activity of preparation and oral exhibition of short and specific topics affine to these subjects. Therefore, the teacher has a relevant and consultant role of the selected topic and his function is to guide and supervise the work, incorporating many times the previous works done in other courses, as a research tutor or more experienced student. Then, in this work, we show the academic results, grade of interactivity developed in these collaborative tasks, statistics and the satisfaction grade shown by our post-graduate students.

An examination of learning processes during critical incident training: Implications for the development of adaptable trainees

Three experiments are reported that examined the process by which trainees learn decision-making skills during a critical incident training program. Formal theories of category learning were used to identify two processes that may be responsible for the acquisition of decision-making skills: rule learning and exemplar learning. Experiments I and 2 used the process dissociation procedure (L. L. Jacoby, 1998) to evaluate the contribution of these processes to performance. The results suggest that trainees used a mixture of rule and exemplar learning. Furthermore, these learning processes were influenced by different aspects of training structure and design. The goal of Experiment 3 was to develop training techniques that enable trainees to use a rule adaptively. Trainees were tested on cases that represented exceptions to the rule. Unexpectedly, the results suggest that providing general instruction regarding the kinds of conditions in which a decision rule does not apply caused them to fixate on the specific conditions mentioned and impaired their ability to identify other conditions in which the rule might not apply. The theoretical, methodological, and practical implications of the results are discussed.