Refine
Year of publication
Document Type
- Article (51) (remove)
Keywords
- Security (3)
- e-learning (3)
- Digitale Bildung (2)
- Energy (2)
- Kursdesign (2)
- MOOC (2)
- Micro Degree (2)
- Online-Lehre (2)
- Onlinekurs (2)
- Onlinekurs-Produktion (2)
Generative multi-adversarial network for striking the right balance in abdominal image segmentation
(2020)
Purpose: The identification of abnormalities that are relatively rare within otherwise normal anatomy is a major challenge for deep learning in the semantic segmentation of medical images. The small number of samples of the minority classes in the training data makes the learning of optimal classification challenging, while the more frequently occurring samples of the majority class hamper the generalization of the classification boundary between infrequently occurring target objects and classes. In this paper, we developed a novel generative multi-adversarial network, called Ensemble-GAN, for mitigating this class imbalance problem in the semantic segmentation of abdominal images. Method: The Ensemble-GAN framework is composed of a single-generator and a multi-discriminator variant for handling the class imbalance problem to provide a better generalization than existing approaches. The ensemble model aggregates the estimates of multiple models by training from different initializations and losses from various subsets of the training data. The single generator network analyzes the input image as a condition to predict a corresponding semantic segmentation image by use of feedback from the ensemble of discriminator networks. To evaluate the framework, we trained our framework on two public datasets, with different imbalance ratios and imaging modalities: the Chaos 2019 and the LiTS 2017. Result: In terms of the F1 score, the accuracies of the semantic segmentation of healthy spleen, liver, and left and right kidneys were 0.93, 0.96, 0.90 and 0.94, respectively. The overall F1 scores for simultaneous segmentation of the lesions and liver were 0.83 and 0.94, respectively. Conclusion: The proposed Ensemble-GAN framework demonstrated outstanding performance in the semantic segmentation of medical images in comparison with other approaches on popular abdominal imaging benchmarks. The Ensemble-GAN has the potential to segment abdominal images more accurately than human experts.
TransPipe
(2021)
Online learning environments, such as Massive Open Online Courses (MOOCs), often rely on videos as a major component to convey knowledge. However, these videos exclude potential participants who do not understand the lecturer’s language, regardless of whether that is due to language unfamiliarity or aural handicaps. Subtitles and/or interactive transcripts solve this issue, ease navigation based on the content, and enable indexing and retrieval by search engines. Although there are several automated speech-to-text converters and translation tools, their quality varies and the process of integrating them can be quite tedious. Thus, in practice, many videos on MOOC platforms only receive subtitles after the course is already finished (if at all) due to a lack of resources. This work describes an approach to tackle this issue by providing a dedicated tool, which is closing this gap between MOOC platforms and transcription and translation tools and offering a simple workflow that can easily be handled by users with a less technical background. The proposed method is designed and evaluated by qualitative interviews with three major MOOC providers.
Generating a novel and descriptive caption of an image is drawing increasing interests in computer vision, natural language processing, and multimedia communities. In this work, we propose an end-to-end trainable deep bidirectional LSTM (Bi-LSTM (Long Short-Term Memory)) model to address the problem. By combining a deep convolutional neural network (CNN) and two separate LSTM networks, our model is capable of learning long-term visual-language interactions by making use of history and future context information at high-level semantic space. We also explore deep multimodal bidirectional models, in which we increase the depth of nonlinearity transition in different ways to learn hierarchical visual-language embeddings. Data augmentation techniques such as multi-crop, multi-scale, and vertical mirror are proposed to prevent over-fitting in training deep models. To understand how our models "translate" image to sentence, we visualize and qualitatively analyze the evolution of Bi-LSTM internal states over time. The effectiveness and generality of proposed models are evaluated on four benchmark datasets: Flickr8K, Flickr30K, MSCOCO, and Pascal1K datasets. We demonstrate that Bi-LSTM models achieve highly competitive performance on both caption generation and image-sentence retrieval even without integrating an additional mechanism (e.g., object detection, attention model). Our experiments also prove that multi-task learning is beneficial to increase model generality and gain performance. We also demonstrate the performance of transfer learning of the Bi-LSTM model significantly outperforms previous methods on the Pascal1K dataset.
Multimodal representation learning has gained increasing importance in various real-world multimedia applications. Most previous approaches focused on exploring inter-modal correlation by learning a common or intermediate space in a conventional way, e.g. Canonical Correlation Analysis (CCA). These works neglected the exploration of fusing multiple modalities at higher semantic level. In this paper, inspired by the success of deep networks in multimedia computing, we propose a novel unified deep neural framework for multimodal representation learning. To capture the high-level semantic correlations across modalities, we adopted deep learning feature as image representation and topic feature as text representation respectively. In joint model learning, a 5-layer neural network is designed and enforced with a supervised pre-training in the first 3 layers for intra-modal regularization. The extensive experiments on benchmark Wikipedia and MIR Flickr 25K datasets show that our approach achieves state-of-the-art results compare to both shallow and deep models in multimodal and cross-modal retrieval.
Effective classroom management is considered a key criterion to making classrooms effective learning environments. Supporting classroom orchestration—the teacher-centric real-time management of classroom activities—is central to achieving effective classroom management. However, the multi-faceted nature of classroom orchestration, its complexity, and general classroom constraints such as time, present challenges for the effective management of the modern-day classroom environment. Though effective, most existing approaches for overcoming orchestration challenges, such as Google Classroom, are arguably ad hoc. We argue that streamlined technology-driven orchestration can be attained through the use of an orchestration workbench, potentially making educators more effective within formal learning environments. Early supporting evidence, from a study involving the use of a prototype orchestration tool, demonstrates the feasibility of organised orchestration and its potential to improve students' learning experience.
Coordinated sampled listening (CSL) is a standardized medium access control protocol for IEEE 80215.4 networks. Unfortunately, CSL comes without any protection against so-called denial-of-sleep attacks. Such attacks deprive energy-constrained devices of entering low-power sleep modes, thereby draining their charge. Repercussions of denial-of-sleep attacks include long outages, violated quality-of-service guarantees, and reduced customer satisfaction. However, while CSL has no built-in denial-of-sleep defenses, there already exist denial-of-sleep defenses for a predecessor of CSL, namely ContikiMAC. In this paper, we make two main contributions. First, motivated by the fact that CSL has many advantages over ContikiMAC, we tailor the existing denial-of-sleep defenses for ContikiMAC to CSL. Second, we propose several security enhancements to these existing denial-of-sleep defenses. In effect, our denial-of-sleep defenses for CSL mitigate denial-of-sleep attacks significantly better, as well as protect against a larger range of denial-of-sleep attacks than the existing denial-of-sleep defenses for ContikiMAC. We show the soundness of our denial-of-sleep defenses for CSL both analytically, as well as empirically using a whole new implementation of CSL. (C) 2018 Elsevier B.V. All rights reserved.
A Fuzzy Rule-Based Model for Remote Monitoring of Preterm in the Intensive Care Unit of Hospitals
(2019)
The use of Remote patient monitoring (RPM) systems to monitor critically ill patients in the Intensive Care Unit (ICU) has enabled quality and real-time healthcare management. Fuzzy logic as an approach to designing RPM systems provides a means for encapsulating the subjective decision-making process of medical experts in an algorithm suitable for computer implementation. In this paper, a remote monitoring system for preterm in neonatal ICU incubators is modeled and simulated. The model was designed with 4 input variables (body temperature, heart rate, respiratory rate, and oxygen level saturation), and 1 output variable (action performed represented as ACT). ACT decides whether-an alert is generated or not and also determines the message displayed when a notification is required. ACT classifies the clinical priority of the monitored preterm into 5 different fields: code blue, code red, code yellow, code green, and-code black. The model was simulated using a fuzzy logic toolbox of MATLAB R2015A. About 216 IF_THEN rules were formulated to monitor the inputs data fed into the model. The performance of the model was evaluated using-the confusion matrix to determine the model’s accuracy, precision, sensitivity, specificity, and false alarm rate. The-experimental results obtained shows that the fuzzy-based system is capable of producing satisfactory results when used for monitoring and classifying the clinical statuses of neonates in ICU incubators.
Electronic health is one of the most popular applications of information and communication technologies and it has contributed immensely to health delivery through the provision of quality health service and ubiquitous access at a lower cost. Even though this mode of health service is increasingly becoming known or used in developing nations, these countries are faced with a myriad of challenges when implementing and deploying e-health services on both small and large scale. It is estimated that the Africa population alone carries the highest percentage of the world’s global diseases despite its certain level of e-health adoption. This paper aims at analyzing the progress so far and the current state of e-health in developing countries particularly Africa and propose a framework for further improvement.
The relevance of identity data leaks on the Internet is more present than ever. Almost every week we read about leakage of databases with more than a million users in the news. Smaller but not less dangerous leaks happen even multiple times a day. The public availability of such leaked data is a major threat to the victims, but also creates the opportunity to learn not only about security of service providers but also the behavior of users when choosing passwords. Our goal is to analyze this data and generate knowledge that can be used to increase security awareness and security, respectively. This paper presents a novel approach to the processing and analysis of a vast majority of bigger and smaller leaks. We evolved from a semi-manual to a fully automated process that requires a minimum of human interaction. Our contribution is the concept and a prototype implementation of a leak processing workflow that includes the extraction of digital identities from structured and unstructured leak-files, the identification of hash routines and a quality control to ensure leak authenticity. By making use of parallel and distributed programming, we are able to make leaks almost immediately available for analysis and notification after they have been published. Based on the data collected, this paper reveals how easy it is for criminals to collect lots of passwords, which are plain text or only weakly hashed. We publish those results and hope to increase not only security awareness of Internet users but also security on a technical level on the service provider side.