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.

Financial Information Mediation: A Case Study of Standards Integration for Electronic Bill Presentment and Payment Using the COIN Mediation Technology

Each player in the financial industry, each bank, stock exchange, government agency, or insurance company operates its own financial information system or systems. By its very nature, financial information, like the money that it represents, changes hands. Therefore the interoperation of financial information systems is the cornerstone of the financial services they support. E-services frameworks such as web services are an unprecedented opportunity for the flexible interoperation of financial systems. Naturally the critical economic role and the complexity of financial information led to the development of various standards. Yet standards alone are not the panacea: different groups of players use different standards or different interpretations of the same standard. We believe that the solution lies in the convergence of flexible E-services such as web-services and semantically rich meta-data as promised by the semantic Web; then a mediation architecture can be used for the documentation, identification, and resolution of semantic conflicts arising from the interoperation of heterogeneous financial services. In this paper we illustrate the nature of the problem in the Electronic Bill Presentment and Payment (EBPP) industry and the viability of the solution we propose. We describe and analyze the integration of services using four different formats: the IFX, OFX and SWIFT standards, and an example proprietary format. To accomplish this integration we use the COntext INterchange (COIN) framework. The COIN architecture leverages a model of sources and receivers’ contexts in reference to a rich domain model or ontology for the description and resolution of semantic heterogeneity.

Recommender System Based on Consumer Product Reviews

Consumer reviews, opinions and shared experiences in the use of a product is a powerful source of information about consumer preferences that can be used in recommender systems. Despite the importance and value of such information, there is no comprehensive mechanism that formalizes the opinions selection and retrieval process and the utilization of retrieved opinions due to the difficulty of extracting information from text data. In this paper, a new recommender system that is built on consumer product reviews is proposed. A prioritizing mechanism is developed for the system. The proposed approach is illustrated using the case study of a recommender system for digital cameras

Java computer modelling of the Earth System

The JModel suite consists of a number of models of aspects of the Earth System. The Java programmes model in detail aspects of the cycles of some major biogeochemical elements that exemplify the range of geochemical processes in marine environments.

Evaluation of the effectiveness of an organization in the field of brand perception using a computer agent-based simulation system

Este proyecto de investigación busca usar un sistema de cómputo basado en modelación por agentes para medir la percepción de marca de una organización en una población heterogénea. Se espera proporcionar información que permita dar soluciones a una organización acerca del comportamiento de sus consumidores y la asociada percepción de marca. El propósito de este sistema es el de modelar el proceso de percepción-razonamiento-acción para simular un proceso de razonamiento como el resultado de una acumulación de percepciones que resultan en las acciones del consumidor. Este resultado definirá la aceptación de marca o el rechazo del consumidor hacia la empresa. Se realizó un proceso de recolección información acerca de una organización específica en el campo de marketing. Después de compilar y procesar la información obtenida de la empresa, el análisis de la percepción de marca es aplicado mediante procesos de simulación. Los resultados del experimento son emitidos a la organización mediante un informe basado en conclusiones y recomendaciones a nivel de marketing para mejorar la percepción de marca por parte de los consumidores.

La enseñanza y aprendizaje del análisis matemático haciendo uso de CAS (computer algebra system).

Se presentan dos investigaciones sobre la enseñanza y el aprendizaje de la integral definida y la integral impropia. Se destacan los aspectos relacionados con el uso de los CAS (Computer Algebra System) Derive y Maple. Se hace incapié en el papel que ha jugado cada uno de ellos en la investigación.

The impact of a closed-loop electronic prescribing and administration system on prescribing errors, administration errors and staff time: a before-and-after study

Objectives: To assess the impact of a closed-loop electronic prescribing, automated dispensing, barcode patient identification and electronic medication administration record (EMAR) system on prescribing and administration errors, confirmation of patient identity before administration, and staff time. Design, setting and participants: Before-and-after study in a surgical ward of a teaching hospital, involving patients and staff of that ward. Intervention: Closed-loop electronic prescribing, automated dispensing, barcode patient identification and EMAR system. Main outcome measures: Percentage of new medication orders with a prescribing error, percentage of doses with medication administration errors (MAEs) and percentage given without checking patient identity. Time spent prescribing and providing a ward pharmacy service. Nursing time on medication tasks. Results: Prescribing errors were identified in 3.8% of 2450 medication orders pre-intervention and 2.0% of 2353 orders afterwards (p<0.001; χ2 test). MAEs occurred in 7.0% of 1473 non-intravenous doses pre-intervention and 4.3% of 1139 afterwards (p = 0.005; χ2 test). Patient identity was not checked for 82.6% of 1344 doses pre-intervention and 18.9% of 1291 afterwards (p<0.001; χ2 test). Medical staff required 15 s to prescribe a regular inpatient drug pre-intervention and 39 s afterwards (p = 0.03; t test). Time spent providing a ward pharmacy service increased from 68 min to 98 min each weekday (p = 0.001; t test); 22% of drug charts were unavailable pre-intervention. Time per drug administration round decreased from 50 min to 40 min (p = 0.006; t test); nursing time on medication tasks outside of drug rounds increased from 21.1% to 28.7% (p = 0.006; χ2 test). Conclusions: A closed-loop electronic prescribing, dispensing and barcode patient identification system reduced prescribing errors and MAEs, and increased confirmation of patient identity before administration. Time spent on medication-related tasks increased.

The impact of a closed-loop electronic prescribing and automated dispensing system on the ward pharmacist’s time and activities

Objective To assess the impact of a closed-loop electronic prescribing and automated dispensing system on the time spent providing a ward pharmacy service and the activities carried out. Setting Surgical ward, London teaching hospital. Method All data were collected two months pre- and one year post-intervention. First, the ward pharmacist recorded the time taken each day for four weeks. Second, an observational study was conducted over 10 weekdays, using two-dimensional work sampling, to identify the ward pharmacist's activities. Finally, medication orders were examined to identify pharmacists' endorsements that should have been, and were actually, made. Key findings Mean time to provide a weekday ward pharmacy service increased from 1 h 8 min to 1 h 38 min per day (P = 0.001; unpaired t-test). There were significant increases in time spent prescription monitoring, recommending changes in therapy/monitoring, giving advice or information, and non-productive time. There were decreases for supply, looking for charts and checking patients' own drugs. There was an increase in the amount of time spent with medical and pharmacy staff, and with 'self'. Seventy-eight per cent of patients' medication records could be assessed for endorsements pre- and 100% post-intervention. Endorsements were required for 390 (50%) of 787 medication orders pre-intervention and 190 (21%) of 897 afterwards (P < 0.0001; chi-square test). Endorsements were made for 214 (55%) of endorsement opportunities pre-intervention and 57 (30%) afterwards (P < 0.0001; chi-square test). Conclusion The intervention increased the overall time required to provide a ward pharmacy service and changed the types of activity undertaken. Contact time with medical and pharmacy staff increased. There was no significant change in time spent with patients. Fewer pharmacy endorsements were required post-intervention, but a lower percentage were actually made. The findings have important implications for the design, introduction and use of similar systems.

Extending the human nervous system through Internet implants - Experimentation and impact

It is now possible to directly link the human nervous system to a computer and thence onto the Internet. From an electronic and mental viewpoint this means that the Internet becomes an extension of the human nervous system (and vice versa). Such a connection on a regular or mass basis will have far reaching effects for society. In this article the authors discuss their own practical implant self-experimentation, especially insofar as it relates to extending the human nervous system. Trials involving an intercontinental link up are described. As well as technical aspects of the work, social, moral and ethical issues, as perceived by the authors, are weighed against potential technical gains. The authors also look at technical limitations inherent in the co-evolution of Internet implanted individuals as well as the future distribution of intelligence between human and machine.

Numerical precision requirements on the Multiband Ultra-Wideband system for practical consumer electronic devices

This paper discusses the requirements on the numerical precision for a practical Multiband Ultra-Wideband (UWB) consumer electronic solution. To this end we first present the possibilities that UWB has to offer to the consumer electronics market and the possible range of devices. We then show the performance of a model of the UWB baseband system implemented using floating point precision. Then, by simulation we find the minimal numerical precision required to maintain floating-point performance for each of the specific data types and signals present in the UWB baseband. Finally, we present a full description of the numerical requirements for both the transmit and receive components of the UWB baseband. The numerical precision results obtained in this paper can then be used by baseband designers to implement cost effective UWB systems using System-on-Chip (SoC), FPGA and ASIC technology solutions biased toward the competitive consumer electronics market(1).

A CAD-based computer vision system

A vision system for recognizing rigid and articulated three-dimensional objects in two-dimensional images is described. Geometrical models are extracted from a commercial computer aided design package. The models are then augmented with appearance and functional information which improves the system's hypothesis generation, hypothesis verification, and pose refinement. Significant advantages over existing CAD-based vision systems, which utilize only information available in the CAD system, are realized. Examples show the system recognizing, locating, and tracking a variety of objects in a robot work-cell and in natural scenes.

Determining the Trustworthiness of New Electronic Contracts

Expressing contractual agreements electronically potentially allows agents to automatically perform functions surrounding contract use: establishment, fulfilment, renegotiation etc. For such automation to be used for real business concerns, there needs to be a high level of trust in the agent-based system. While there has been much research on simulating trust between agents, there are areas where such trust is harder to establish. In particular, contract proposals may come from parties that an agent has had no prior interaction with and, in competitive business-to-business environments, little reputation information may be available. In human practice, trust in a proposed contract is determined in part from the content of the proposal itself, and the similarity of the content to that of prior contracts, executed to varying degrees of success. In this paper, we argue that such analysis is also appropriate in automated systems, and to provide it we need systems to record salient details of prior contract use and algorithms for assessing proposals on their content. We use provenance technology to provide the former and detail algorithms for measuring contract success and similarity for the latter, applying them to an aerospace case study.