73 research outputs found
What Explains the Canada-US ICT Investment Intensity Gap?
No full textIt is widely recognized that machinery and equipment investment intensity is less in Canada than in the United States. What is less well know is that it is information and communications technology (ICT) investment that largely accounts for this gap. The author documents trends in ICT investment in both Canada and the United States and attempts to explain why ICT investment per worker in the Canadian business sector in 2004 was only 45 per cent of that in the US business sector. While no definitive explanation emerges, among the factors he identifies as playing a role are industrial structure, firm size distribution of employment, the price of labour compared to ICT investment goods, and the underestimation of ICT investment in official statistics.Machinery and equipment investment, information and communications technology, ICT, Investment gap, Business sector, Industrial structure, Firm size
Estimating ambient visibility in the presence of fog: a deep convolutional neural network approach
No full textNext-generation intelligent transportation systems are based on the acquisition of ambient data that influence traffic flow and safety. Among these, is the ambient visibility range whose estimation, in the presence of fog, is extremely useful for next-generation intelligent transportation systems. However, existing camera-based approaches are based on "engineered features" extraction methods that use computer algorithms and procedures from the image processing field. In this contribution, a novel approach to estimate visibility range under foggy weather conditions is proposed which is based on "learned features" instead. More precisely, we use AlexNet deep convolutional neural network (DCNN), trained with raw image data, for feature extraction and a support vector machine (SVM) for visibility range estimation. Our quantitative analysis showed that the proposed approach is very promising in estimating the visibility range with very good accuracy. The proposed solution can pave the way towards intelligent driveway assistance systems to enhance awareness of driving weather conditions and hence mitigate the safety risks emanating from fog-induced low visibility conditions.This research was financially supported by Zayed University Cluster Research Grant No. R17075.Kamoun, F (reprint author), ESPRIT Sch Engn, Tunis, Tunisia.
[email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Multi-label classification of a real-world image dataset
No full textDeep neural networks have shown increasing performance in image classification recent years. However, most of the methods developed in research are created, trained and compared on standard, general image collections. While some of such datasets contain realistic pictures with multiple objects and interaction between them, the selection of pictures and categories is broad and designed for the training of automatic classification systems. Therefore, questions remain on how modern image classification approaches can be applied to real-world datasets. What challenges can arise in the implementation of systems based on such datasets and how can they be solved? What level of classification performance can be expected? This study looks into these questions and gives insights on building such classification systems for real-world image collections.
The Norwegian News Agency provided the author with a unique labeled dataset of one million images for the purpose of this research. The study consists of a trial and main experiment. The goal of the trial experiments was to learn necessary tools, test classification system implementation, training and testing pipelines as well as different architectural design choices. These set of experiments were performed on a subset of the original image collection with a limited number of categories. Insights from these experiments were further used to perform model training on a much broader set of manually selected categories. Two network architectures were employed and compared in the research: CaffeNet and GoogleNet.
Results from the study suggest a big potential of using pre-trained convolutional neural networks in solving the task of multi-label image classification on a real-world dataset. However, the study also highlights a number of challenges connected with developing such system in a realistic environment, including:
A unique set of categories which requires either to train a new model or to reuse a pre-trained neural network with adjustments in the architecture to solve classification problems.
Difficulties in category tree transformation connected with the decision of which categories include, exclude and merge.
Categories which are unsuitable for automatic training purposes including contextual, abstract and ambiguous labels.
Contextual images which can not be classified only by visual features and require additional information.
Issues connected with the classification of objects and people located in the background of the pictures.
Duplicates of images in the dataset which can cause issues for both training and testing processes.
The research gives further insights on the challenges mentioned, discusses their impact and possible solutions to them. The main limitation of the study is that only
one dataset was employed in the research. Therefore, results are considered likely to be more generalizable to datasets similar to the one used in the study
Mobile NFC Services: Adoption Factors and a Typology of Business Models
No full textThe integration of Near Field Communication (NFC) into mobile devices has recently emerged as a disruptive innovation and a strong enabler of a wide range of new mobile applications and services. Yet, despite this great potential, the widespread adoption of mobile NFC services remains restrained. This chapter investigates the success factors that are contributing towards the proliferation of mobile NFC services. It also presents an in-depth analysis of the key hurdles standing in the way of full NFC commercialization, with the caveat that disagreement about the proper business model among the key ecosystem players is currently the major adoption restraint. The chapter articulates the vision of a cooperative model that can enable the sharing of services, infrastructure, cost, and revenues among various NFC ecosystem players. The author also adopts a four-tier classification approach to categorize NFC business models into a number of typologies. Some recommendations for future research are also provided
Cardiac hydatid cyst revealed by ventricular tachycardia
No full textAbstractHydatid disease is a human parasitic infestation caused by the larval stage of Echinococcus Granulosus. The liver and the lungs are the most common locations. Cardiac involvement is rare and accounts for 0.5–2% of all hydatid disease. We report an unusual presentation of cardiac hydatid cyst revealed by ventricular tachycardia in a patient with a history of cerebral hydatid cyst
An unusual case of Behçet's disease presenting with postpartum ovarian iliac vein thrombosis and pulmonary embolism
No full textAbstract Thrombosis of the ovarian vein is a rare complication which arises classically in the postpartum. We report a case of 24-year-old woman with a history of Behçet's disease, who presented with pelvic and thoracic pain, tachycardia, dyspnea and fever occurring 2 weeks after delivery. Computed tomography revealed an ascending thrombosis of the iliac and right ovarian veins complicated by bilateral pulmonary embolism. The patient responded well to the combination of anticoagulants and immunosuppressive agents. Behçet's disease should also be considered as an etiologic factor for ovarian vein thrombosis.</p
Multi-label classification of a real-world image dataset
No full textDeep neural networks have shown increasing performance in image classification recent years. However, most of the methods developed in research are created, trained and compared on standard, general image collections. While some of such datasets contain realistic pictures with multiple objects and interaction between them, the selection of pictures and categories is broad and designed for the training of automatic classification systems. Therefore, questions remain on how modern image classification approaches can be applied to real-world datasets. What challenges can arise in the implementation of systems based on such datasets and how can they be solved? What level of classification performance can be expected? This study looks into these questions and gives insights on building such classification systems for real-world image collections.
The Norwegian News Agency provided the author with a unique labeled dataset of one million images for the purpose of this research. The study consists of a trial and main experiment. The goal of the trial experiments was to learn necessary tools, test classification system implementation, training and testing pipelines as well as different architectural design choices. These set of experiments were performed on a subset of the original image collection with a limited number of categories. Insights from these experiments were further used to perform model training on a much broader set of manually selected categories. Two network architectures were employed and compared in the research: CaffeNet and GoogleNet.
Results from the study suggest a big potential of using pre-trained convolutional neural networks in solving the task of multi-label image classification on a real-world dataset. However, the study also highlights a number of challenges connected with developing such system in a realistic environment, including:
A unique set of categories which requires either to train a new model or to reuse a pre-trained neural network with adjustments in the architecture to solve classification problems.
Difficulties in category tree transformation connected with the decision of which categories include, exclude and merge.
Categories which are unsuitable for automatic training purposes including contextual, abstract and ambiguous labels.
Contextual images which can not be classified only by visual features and require additional information.
Issues connected with the classification of objects and people located in the background of the pictures.
Duplicates of images in the dataset which can cause issues for both training and testing processes.
The research gives further insights on the challenges mentioned, discusses their impact and possible solutions to them. The main limitation of the study is that only
one dataset was employed in the research. Therefore, results are considered likely to be more generalizable to datasets similar to the one used in the study
Human and Organizational Factors of Healthcare Data Breaches
No full textCopyright © 2014, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. Over the past few years, concerns related to healthcare data privacy have been mounting since healthcare information has become more digitized, distributed and mobile. However, very little is known about the root cause of data breach incidents; making it difficult for healthcare organizations to establish proper security controls and defenses. Through a systematic review and synthesis of data breaches literature, and using databases of earlier reported healthcare data breaches, the authors re-examine and analyze the causal factors behind healthcare data breaches. The authors then use the Swiss Cheese Model (SCM) to shed light on the technical, organizational and human factors of these breaches. The author\u27s research suggests that incorporating the SCM concepts into the healthcare security policies and procedures can assist healthcare providers in assessing the vulnerabilities and risks associated with the maintenance and transmission of protected health information
- …
