Leveraging noisy side information for disentangling of factors of variation in a supervised setting

Ce mémoire est composé de trois articles et présente les résultats de travaux de recherche effectués dans le but d'améliorer les techniques actuelles permettant d'utiliser des données associées à certaines tâches dans le but d'aider à l'entraînement de réseaux de neurones sur une tâche différente. Les deux premiers articles présentent de nouveaux ensembles de données créés pour permettre une meilleure évaluation de ce type de techniques d'apprentissage machine. Le premier article introduit une suite d'ensembles de données pour la tâche de reconnaissance automatique de chiffres écrits à la main. Ces ensembles de données ont été générés à partir d'un ensemble de données déjà existant, MNIST, auquel des nouveaux facteurs de variation ont été ajoutés. Le deuxième article introduit un ensemble de données pour la tâche de reconnaissance automatique d'expressions faciales. Cet ensemble de données est composé d'images de visages qui ont été collectées automatiquement à partir du Web et ensuite étiquetées. Le troisième et dernier article présente deux nouvelles approches, dans le contexte de l'apprentissage multi-tâches, pour tirer avantage de données pour une tâche donnée afin d'améliorer les performances d'un modèle sur une tâche différente. La première approche est une généralisation des neurones Maxout récemment proposées alors que la deuxième consiste en l'application dans un contexte supervisé d'une technique permettant d'inciter des neurones à apprendre des fonctions orthogonales, à l'origine proposée pour utilisation dans un contexte semi-supervisé.

Towards Non Invasive Diagnosis of Scoliosis Using Semi-supervised Learning Approach

In this paper, a new methodology for the prediction of scoliosis curve types from non invasive acquisitions of the back surface of the trunk is proposed. One hundred and fifty-nine scoliosis patients had their back surface acquired in 3D using an optical digitizer. Each surface is then characterized by 45 local measurements of the back surface rotation. Using a semi-supervised algorithm, the classifier is trained with only 32 labeled and 58 unlabeled data. Tested on 69 new samples, the classifier succeeded in classifying correctly 87.0% of the data. After reducing the number of labeled training samples to 12, the behavior of the resulting classifier tends to be similar to the reference case where the classifier is trained only with the maximum number of available labeled data. Moreover, the addition of unlabeled data guided the classifier towards more generalizable boundaries between the classes. Those results provide a proof of feasibility for using a semi-supervised learning algorithm to train a classifier for the prediction of a scoliosis curve type, when only a few training data are labeled. This constitutes a promising clinical finding since it will allow the diagnosis and the follow-up of scoliotic deformities without exposing the patient to X-ray radiations.

OPTIMAL UTILIZATION OF SERVICE FACILITY FOR A k-OUT-OF-n SYSTEM WITH REPAIR BY EXTENDING SERVICE TO EXTERNAL CUSTOMERS IN A RETRIAL QUEUE

In this paper, we study a k-out-of-n system with single server who provides service to external customers also. The system consists of two parts:(i) a main queue consisting of customers (failed components of the k-out-of-n system) and (ii) a pool (of finite capacity M) of external customers together with an orbit for external customers who find the pool full. An external customer who finds the pool full on arrival, joins the orbit with probability and with probability 1− leaves the system forever. An orbital customer, who finds the pool full, at an epoch of repeated attempt, returns to orbit with probability (< 1) and with probability 1 − leaves the system forever. We compute the steady state system size probability. Several performance measures are computed, numerical illustrations are provided.

Influence of Angle of Air Injection and Particles in Bed Hydrodynamics of Swirling Fluidized Bed

The thesis presented here unveils an experimental study of the hydrodynamic characteristics of swirling fluidized bed viz. pressure drop across the distributor and the bed, minimum fluidizing velocity, bed behaviour and angle of air injection. In swirling fluidized bed the air is admitted to the bed at an angle 'Ѳ' to the horizontal. The vertical component of the velocity v sin Ѳ causes fluidization and the horizontal component v cos Ѳ contributes to swirl motion of the bed material.The study was conducted using spherical particles having sizes 3.2 mm, 5.5 mm & 7.4 mm as the bed materials. Each of these particles was made from high density polyethylene, nylon and acetal having relative densities of 0.93, 1.05 and 1.47 respectively.The experiments were conducted using conidour type distributors having four rows of slits. Altogether four distributors having angles of air injection (Φ)- 0°, 5°, 10° & 15° were designed and fabricated for the study. The total number of slits in each distributor was 144. The area of opening was 6220 mm2 making the percentage area of opening to 9.17. But the percentage useful area of opening of the distributor was 96.The experiments on the variation of distributor pressure drop with superficial velocity revealed that the distributor pressure drop decreases with angle of air injection. Investigations related to bed hydrodynamics were conducted using 2.5 kg of bed material. The bed pressure drop measurements were made along the radial direction of the distributor at distances of 60 mm, 90 mm, 120 mm & 150 mm from the centre of the distributor. It was noticed that after attaining minimum fluidizing velocity, the bed pressure drop increases along the radial direction of the distributor. But at a radial distance of 90 mm from the distributor centre, after attaining minimum fluidizing velocity the bed pressure drop remains almost constant. It was also observed that the bed pressure drop varies inversely with particle size as well as particle density.An attempt was made to determine the effect of various parameters on minimum fluidizing velocity. It was noticed that the minimum fluidizing velocity varies directly with angle of air injection (Φ), particle size and particle density.The study on the bed behaviour showed that the superficial velocity required for initiating various bed phenomena (such as swirl motion and separation of particles from the cone at the centre) increase with increase in particle size as well as particle density. It was also observed that the particle size and particle density directly influence the superficial velocity required for various regimes of bed behaviour such as linear variation of bed pressure drop, constant bed pressure drop and sudden increase or decrease in bed pressure drop.Experiments were also performed to study the effect of angle of air injection (Φ). It was noticed that the bed pressure drop decreases with angle of air injection. It was also noticed that the angle of air injection directly influence the superficial velocity required for initiating various bed phenomena as well as the various regimes of bed behaviour.

Dynamic Injection of Scribble Features into Graphical Diagram Editors

Almost everyone sketches. People use sketches day in and day out in many different and heterogeneous fields, to share their thoughts and clarify ambiguous interpretations, for example. The media used to sketch varies from analog tools like flipcharts to digital tools like smartboards. Whereas analog tools are usually affected by insufficient editing capabilities like cut/copy/paste, digital tools greatly support these scenarios. Digital tools can be grouped into informal and formal tools. Informal tools can be understood as simple drawing environments, whereas formal tools offer sophisticated support to create, optimize and validate diagrams of a certain application domain. Most digital formal tools force users to stick to a concrete syntax and editing workflow, limiting the user’s creativity. For that reason, a lot of people first sketch their ideas using the flexibility of analog or digital informal tools. Subsequently, the sketch is "portrayed" in an appropriate digital formal tool. This work presents Scribble, a highly configurable and extensible sketching framework which allows to dynamically inject sketching features into existing graphical diagram editors, based on Eclipse GEF. This allows to combine the flexibility of informal tools with the power of formal tools without any effort. No additional code is required to augment a GEF editor with sophisticated sketching features. Scribble recognizes drawn elements as well as handwritten text and automatically generates the corresponding domain elements. A local training data library is created dynamically by incrementally learning shapes, drawn by the user. Training data can be shared with others using the WebScribble web application which has been created as part of this work.

Ecosystem services and adaptation for agriculture and food security: analysis of two decades (1991–2011) of financing by the Global Environment Facility

Investing in global environmental and adaptation benefits in the context of agriculture and food security initiatives can play an important role in promoting sustainable intensification. This is a priority for the Global Environment Facility (GEF), created in 1992 with a mandate to serve as financial mechanism of several multilateral environmental agreements. To demonstrate the nature and extent of GEF financing, we conducted an assessment of the entire portfolio over a period of two decades (1991–2011) to identify projects with direct links to agriculture and food security. A cohort of 192 projects and programs were identified and used as a basis for analyzing trends in GEF financing. The projects and programs together accounted for a total GEF financing of US$1,086.8 million, and attracted an additional US$6,343.5 million from other sources. The value-added of GEF financing for ecosystem services and resilience in production systems was demonstrated through a diversity of interventions in the projects and programs that utilized US$810.6 million of the total financing. The interventions fall into the following four main categories in accordance with priorities of the GEF: sustainable land management (US$179.3 million), management of agrobiodiversity (US$113.4 million), sustainable fisheries and water resource management (US$379.8 million), and climate change adaptation (US$138.1 million). By aligning GEF priorities with global aspirations for sustainable intensification of production systems, the study shows that it is possible to help developing countries tackle food insecurity while generating global environmental benefits for a healthy and resilient planet.

People Recognition in Image Sequences by Supervised Learning

We describe a system that learns from examples to recognize people in images taken indoors. Images of people are represented by color-based and shape-based features. Recognition is carried out through combinations of Support Vector Machine classifiers (SVMs). Different types of multiclass strategies based on SVMs are explored and compared to k-Nearest Neighbors classifiers (kNNs). The system works in real time and shows high performance rates for people recognition throughout one day.

Object and Scene Classification: what does a Supervised Approach Provide us?

Given a set of images of scenes containing different object categories (e.g. grass, roads) our objective is to discover these objects in each image, and to use this object occurrences to perform a scene classification (e.g. beach scene, mountain scene). We achieve this by using a supervised learning algorithm able to learn with few images to facilitate the user task. We use a probabilistic model to recognise the objects and further we classify the scene based on their object occurrences. Experimental results are shown and evaluated to prove the validity of our proposal. Object recognition performance is compared to the approaches of He et al. (2004) and Marti et al. (2001) using their own datasets. Furthermore an unsupervised method is implemented in order to evaluate the advantages and disadvantages of our supervised classification approach versus an unsupervised one

Noise levels and hearing conservation in a water treatment facility

This paper describes a study assessing the sound levels and noise exposures of a municipal water treatment plant to determine the level of employee noise exposure dosages and to make any necessary recommendations regarding reducing the risk of noise induced hearing loss in employees.

Injection of adult mouse mesenchymal stem cells into the developing chick inner ear

This series of experiments attempted to characterize the abilities of stem cells derived from bone marrow and adipose tissue to integrate into the sensory epithelium of the inner ear and to differentiate into hair cells or neural cell types.

Does a supervised recreation program influence the personality of the deaf?

This paper discusses the effect of play on the personality of hearing impaired children.