Refine
Document Type
- Conference proceeding (222) (remove)
Language
- English (222) (remove)
Is part of the Bibliography
- yes (222)
Institute
- Technik (120)
- Informatik (83)
- ESB Business School (18)
- Texoversum (1)
Publisher
- IEEE (222) (remove)
Human pose estimation (HPE) is integral to scene understanding in numerous safety-critical domains involving human-machine interaction, such as autonomous driving or semi-automated work environments. Avoiding costly mistakes is synonymous with anticipating failure in model predictions, which necessitates meta-judgments on the accuracy of the applied models. Here, we propose a straightforward human pose regression framework to examine the behavior of two established methods for simultaneous aleatoric and epistemic uncertainty estimation: maximum a-posteriori (MAP) estimation with Monte-Carlo variational inference and deep evidential regression (DER). First, we evaluate both approaches on the quality of their predicted variances and whether these truly capture the expected model error. The initial assessment indicates that both methods exhibit the overconfidence issue common in deep probabilistic models. This observation motivates our implementation of an additional recalibration step to extract reliable confidence intervals. We then take a closer look at deep evidential regression, which, to our knowledge, is applied comprehensively for the first time to the HPE problem. Experimental results indicate that DER behaves as expected in challenging and adverse conditions commonly occurring in HPE and that the predicted uncertainties match their purported aleatoric and epistemic sources. Notably, DER achieves smooth uncertainty estimates without the need for a costly sampling step, making it an attractive candidate for uncertainty estimation on resource-limited platforms.
This article proposes several modified quasi Z-source dc/dc boost converters. These can achieve soft-switching by using a clamp-switch network comprised of an active switch and a diode in parallel with a capacitor connected across one of the inductors of the Z-source network. In this way, ringing at the transistor switching node is mitigated, and the voltage at the turn-on of the transistor is reduced. Even a zero voltage switching (ZVS) of the main transistor is possible if the capacitor in the clamp-switch network is adequately chosen. The proposed circuit structure and operating mode are described and validated through simulations and measurements on a low-power prototype.
Non-fungible tokens (NFTs) are unique digital assets that have recently gained significant popularity, particularly in the digital art sector. The success of NFTs and other blockchain-based innovations depends on their ac-acceptance and use by consumers. This study aims to understand the impact of moral values on the acceptance of NFTs. Based on a quantitative survey with over 800 complete responses, the analysis shows that moral aspects of NFTs are indeed important for potential users. However, there is an attitude-behavior gap, as the positive impact of moral values on the intention to use NFTs is not reflected in the actual current usage of NFTs by the respondents. This study contributes to knowledge by providing new empirical data on the acceptance of NFTs and highlighting the role of moral values on the acceptance decision.
Advancing mental health diagnostics: AI-based method for depression detection in patient interviews
(2023)
In this paper, we present a novel artificial intelligence (AI) application for depression detection, using advanced transformer networks to analyse clinical interviews. By incorporating simulated data to enhance traditional datasets, we overcome limitations in data protection and privacy, consequently improving the model’s performance. Our methodology employs BERT-based models, GPT-3.5, and ChatGPT-4, demonstrating state-of-the-art results in detecting depression from linguistic patterns and contextual information that significantly outperform previous approaches. Utilising the DAIC-WOZ and Extended-DAIC datasets, our study showcases the potential of the proposed application in revolutionising mental health care through early depression detection and intervention. Empirical results from various experiments highlight the efficacy of our approach and its suitability for real-world implementation. Furthermore, we acknowledge the ethical, legal, and social implications of AI in mental health diagnostics. Ultimately, our study underscores the transformative potential of AI in mental health diagnostics, paving the way for innovative solutions that can facilitate early intervention and improve patient outcomes.
Gamification has been increasingly applied to software engineering education in the past. The approaches vary from applying game elements on a conceptual phase in the course to using specific tools to engage the students more and support their learning goals. However, existing tools usually have game elements, such as quizzes or challenges, but do not provide a more computer game-like experience. Therefore, we try to raise the level of gamified learning experience to another level by proposing Gamify-IT. Gamify-IT is a Unity- and web-based game platform intended to help students learn software engineering. It follows an immersive role-play game characteristic where the students explore a world, find and solve minigames and clear dungeons with SE tasks. Lecturers can configure the worlds, e.g., to add content hints. Furthermore, they can add and configure minigames and dungeons to include exercises in a fully gamified way. Thereby, they customize their course in Gamify-IT to adapt the world very precisely to other materials such as lectures or exercises. Results of an evaluation of our initial prototype show that (i) students like to engage with the platform, (ii) students are motivated to learn when using Gamify-IT, and (iii) the minigames support students in understanding the learning objectives.
Analog integrated circuit sizing still relies heavily on human expert knowledge as previous automation approaches have not found wide-spread acceptance in industry. One strand, the optimization-based automation, is often discarded due to inflated constraining setups, infeasible results or excessive run times. To address these deficits, this work proposes a alternative optimization flow featuring a designer’s intuition for feasible design spaces through integration of expert knowledge based on the gm/ID-method. Moreover, the extensive run times of simulation-based optimization flows are overcome by incorporating computationally efficient machine learning methods. Neural network surrogate models predicting eleven performance parameters increase the evaluation speed by 3 400× on average compared to a simulator. Additionally, they enable the use of optimization algorithms dependent on automatic differentiation, that would otherwise be unavailable in this field. First, an up to 4× more efficient way for sampling training data based on the aforementioned space is detailed. After presenting the architecture and training effort regarding the surrogate models, they are employed as part of the objective function for sizing three operational amplifiers with three different optimization algorithms. Additionally, the benefits of using the gm/ID-method become evident when considering technology migration, as previously found solutions may be reused for other technologies.
Measuring cardiorespiratory parameters in sleep, using non-contact sensors and the Ballistocardiography technique has received much attention due to the low-cost, unobtrusive, and non-invasive method. Designing a user-friendly, simple-to-use, and easy-to-deployment preserving less error-prone remains open and challenging due to the complex morphology of the signal. In this work, using four forcesensitive resistor sensors, we conducted a study by designing four distributions of sensors, in order to simplify the complexity of the system by identifying the region of interest for heartbeat and respiration measurement. The sensors are deployed under the mattress and attached to the bed frame without any interference with the subjects. The four distributions are combined in two linear horizontal, one linear vertical, and one square, covering the influencing region in cardiorespiratory activities. We recruited 4 subjects and acquired data in four regular sleeping positions, each for a duration of 80 seconds. The signal processing was performed using discrete wavelet transform bior 3.9 and smooth level of 4 as well as bandpass filtering. The results indicate that we have achieved the mean absolute error of 2.35 and 4.34 for respiration and heartbeat, respectively. The results recommend the efficiency of a triangleshaped structure of three sensors for measuring heartbeat and respiration parameters in all four regular sleeping positions.
This research evaluates current measurement scales for ambidexterity and proposes a new approach for the measurement of this important construct. We argue that current measurement approaches may be unsuitable to capture the concept of ambidexterity. Through a systematic scale development process, we derive a measurement scale with dual items that simultaneously refer to both dimensions, exploitation and exploration, thus reflecting the true nature of ambidexterity. An extensive pre-test with 39 executives suggests that our scale is suitable for capturing ambidexterity. Our measurement model enhances conceptual clarity of ambidexterity and can serve as a base for future investigations of the concept.
Most Question-answering (QA) systems rely on training data to reach their optimal performance. However, acquiring training data for supervised systems is both time-consuming and resource-intensive. To address this, in this paper, we propose TFCSG, an unsupervised similar question retrieval approach that leverages pre-trained language models and multi-task learning. Firstly, topic keywords in question sentences are extracted sequentially based on a latent topic-filtering algorithm to construct unsupervised training corpus data. Then, the multi-task learning method is used to build the question retrieval model. There are three tasks designed. The first is a short sentence contrastive learning task. The second is the question sentence and its corresponding topic sequence similarity judgment task. The third is using question sentences to generate their corresponding topic sequence task. The three tasks are used to train the language model in parallel. Finally, similar questions are obtained by calculating the cosine similarity between sentence vectors. The comparison experiment on public question datasets that TFCSG outperforms the comparative unsupervised baseline method. And there is no need for manual marking, which greatly saves human resources.
Impact of a large distribution network on radiation characteristics of planar spiral antenna arrays
(2023)
Designing antenna arrays with a central feed point has gained ground in the antenna technique. This approach, which is usually applied because of manufacturing costs, is difficult to achieve and leads to a large feeding network. The impact of which is numerically investigated in the present work. Upon comparing three different antennas, it is shown that the enlargement of the feed strongly affects the antenna's overall dimensions and the antenna's radiation characteristics. The antenna with the plug-in solution is not only small in size but also performs better compared to antennas with a central feed point. Considering the high effort in designing the feed network with a central point and the influence of the resulting enlarged network on the dimensions and radiation characteristics of the antenna, the cost saving in production can be put into perspective.
The 17 SDGs, as agreed upon by the international community, are designed to be implemented across all levels of human activity. Alongside the level of international politics, this also includes the local levels, national politics, wider society, and the economic sphere. Many channels are called on to further implementation, including the transfer of technology to developing and emerging countries. As the patent holders, this must include the active participation of companies. While the literature examines the important role of technology transfer in North-South business-to-business (B2B) partnerships, studies on the technology transfer between European and African companies are scarce. Therefore, in this study we use original data from 26 interviews conducted with managers engaged in sales partnerships between German manufacturers and their distributors in African markets to examine the existence and forms of technology transfer. We find that training and marketing excellence are the predominant forms of technology transfer and based on that suggest a refinement of established frameworks on B2B technology transfer.
OpenAPI, WADL, RAML, and API Blueprint are popular formats for documenting Web APIs. Although these formats are in general both human and machine-readable, only the part of the format describing the syntax of a Web API is machine-understandable. Descriptions, which explain the meaning and purpose of Web API elements, are embedded as natural language text snippets into documents and target human readers but not machines. To enable machines to read and process these state-of-practice Web API documentation, we propose a Transformer model that solves the generic task of identifying a Web API element within a syntax structure that matches a natural language query. For our first prototype, we focus on the Web API integration task of matching output with input parameters and fined-tuned a pre-trained CodeBERT model to the downstream task of question answering with samples from 2,321 OpenAPI documentation. We formulate the original question answering problem as a multiple choice task: given a semantic natural language description of an output parameter (question) and the syntax of the input schema (paragraph), the model chooses the input parameter (answer) in the schema that best matches the description. The paper describes the data preparation, tokenization, and fine-tuning process as well as discusses possible applications of our model as part of a recommender system. Furthermore, we evaluate the generalizability and the robustness of our fine-tuned model, with the result that it achieves an accuracy of 81.46% correctly chosen parameters.
We present the results of an extensive characterization of the performance and stability of a third-order continuous-time delta-sigma modulator with active coefficient error compensation. Using our previously published coefficient tuning technique, process variation induced R-C time-constant (TC) errors in the forward signal path can be compensated indirectly using continuously tunable DACs in the feedback path. To validate our technique experimentally with a range of real TC variations, we designed a modulator with discretely configurable integration capacitor arrays in a 0.35-μm CMOS process. We configured the capacitors of the fabricated device for a range of total TC variations from -28.4 % to +19.3 % and measured the signal-to-noise ratio (SNR) as a function of the input amplitude before and after compensating the variations electrically using the feedback DACs. The results show that our tuning technique is capable of restoring the desired nominal modulator performance over the entire parameter variation range, including the system’s nominal maximum stable amplitude (MSA).
Modern wide bandgap power devices promise higher power conversion performance if the device can be operated reliably. As switching speed increases, the effects of parasitic ringing become more prominent, causing potentially damaging overvoltages during device turn-off. Estimating the expected additional voltage caused by such ringing enables more reliable designs. In this paper, we present an analytical expression to calculate the expected overvoltage caused by parasitic ringing based on parasitic element values and operating point parameters. Simulations and measurements confirm that the expression can be used to find the smallest rise time of the switches’ drain-source voltage for minimum overvoltage. The given expression also allows the prediction of the trade off overvoltage amplitude in case of faster required rise times.
Large critical systems, such as those created in the space domain, are usually developed by a large number of organizations and, furthermore, they have to comply with standards. Yet, the different stakeholders often do not have a common understanding of the needed quality of requirements specifications. Achieving such a common understanding is a laborious process that is currently not sufficiently supported. Moreover, such a common understanding must be aligned with the standards. In this paper, we present an approach that can be used to align the different stakeholder perceptions regarding the quality of requirements specifications. Existing quality models for requirements specifications are analyzed for equivalences, and transferred into a common representation, the so-called Aligned Quality Map (AQM). Furthermore, a process is defined that supports the alignment of different stakeholder perspectives with regard to the quality of requirements specifications using AQM, which is validated in a case study in the context of European space projects. AQM has been created and populated with an initial set of quality models. It is designed in such way that it can be extended to include further quality models. The case study has shown that an alignment of different stakeholder perspectives and the quality model of the European Cooperation for Space Standardization using AQM is feasible. The approach allows for aligning different stakeholder perspectives for a common understanding of the quality of requirements specifications in the context of standards. Furthermore, AQM supports the assessment of requirements specifications.
Software development teams have to face stress caused by deadlines, staff turnover, or individual differences in commitment, expertise, and time zones. While students are typically taught the theory of software project management, their exposure to such stress factors is usually limited. However, preparing students for the stress they will have to endure once they work in project teams is important for their own sake, as well as for the sake of team performance in the face of stress. Team performance has been linked to the diversity of software development teams, but little is known about how diversity influences the stress experienced in teams. In order to shed light on this aspect, we provided students with the opportunity to self-experience the basics of project management in self-organizing teams, and studied the impact of six diversity dimensions on team performance, coping with stressors, and positive perceived learning effects. Three controlled experiments at two universities with a total of 65 participants suggest that the social background impacts the perceived stressors the most, while age and work experience have the highest impact on perceived learnings. Most diversity dimensions have a medium correlation with the quality of work, yet no significant relation to the team performance. This lays the foundation to improve students’ training for software engineering teamwork based on their diversity-related needs and to create diversity-sensitive awareness among educators, employers and researchers.
We propose a novel technique to compensate the effects of R-C / gm-C time-constant (TC) errors due to process variation in continuous-time delta-sigma modulators. Local TC error compensation factors are shifted around in the modulator loop to positions where they can be implemented efficiently with tunable circuit structures, such as current-steering digital-to-analog converters (DAC). This approach constitutes an alternative or supplement to existing compensation techniques, including capacitor or gm tuning. We apply the proposed technique to a third-order, single-bit, low-pass continuous-time delta-sigma modulator in cascaded integrator feedback structure. A feedback path tuning scheme is derived analytically and confirmed numerically using behavioral simulations. The modulator circuit was implemented in a 0.35-μm CMOS process using an active feedback coefficient tuning structure based on current-steering DACs. Post-layout simulations show that with this tuning structure, constant performance and stable operation can be obtained over a wide range of TC variation.
Evaluation of human-robot order picking systems considering the evolution of object detection
(2022)
The automation of intralogistic processes is a major trend, but order picking, one of the core and most cost-intensive tasks in this field, remains mostly manual due to the flexibility required during picking. Reacting to its hard physical and ergonomic strain, the automation of this process is however highly relevant. Robotic picking system would enable the automation of this process from a technical point of view, but the necessity for the system to evolve in time, due to dynamics of logistic environments, faces operations with new challenges that are hardly treated in literature. This unknown scares potential investors, hindering the application of technically feasible solutions. In this paper, a model for the evaluation of the additional cost of training of automated systems during operations is presented, that also considers the savings enabled by the system after its evolution. The proposed approach, that considers different parameters such as capacity, ergonomics and cost, is validated with a case study and discussed.
According to several surveys and statistics, the great majority of companies previously not accustomed to automation are piloting solutions to automate business processes. Those accustomed to automation also attempt to introduce more of it, focusing on automation-unfriendly processes that remained manual. However, when the decision on what and whether to automate is not trivial for evident reasons, even industry leaders may get stuck on an overwhelming question: where to begin automating? The question remains too often unanswered as state-of-the-art methods fail to consider the whole picture. This paper introduces a holistic approach to the decision-making for investments in automation. The method supports the iterative analysis and evaluation of operative processes, providing tools for a quantitative approach to the decision-making. Thanks to the method, a large pool of processes can be first considered and then filtered out in order to select the one that yields the best value for the automation in the specific context. After introducing the method, a case study is reported for validation before the discussion.
Compared to the automotive sector, where automation is the rule, in many other less standardized sectors automation is still the exception. This could soon hurt the productivity of industrialized countries, where the unemployment is low and the population is aging. Phenomena like the recent downfall in productivity, due to lockdowns and social distancing for prevention of health hazards during the COVID19 pandemic, only add to the problem. For these reasons, the relevance, motivation and intention for more automation in less standardized sectors has probably never been higher. However, available statistics say that providers and users of technologies struggle to bring more automation into action in automation-unfriendly sectors. In this paper, we present a decision support method for investment in automation that tackles the problem: the STIC analysis. The method takes a holistic and quantitative approach tying together technological, context-related and economic input parameters and synthetizing them in a final economic indicator. Thanks to the modelling of such parameters, it is possible to gain sensibility on the technological and/or process adjustments that would have the highest impact on the efficiency of the automation, thereby delivering value for both technology users and technology providers.