Refine
Year of publication
- 2018 (101) (remove)
Document Type
- Conference Proceeding (64)
- Article (24)
- Part of a Book (6)
- Doctoral Thesis (5)
- Patent (1)
- Working Paper (1)
Language
- English (101) (remove)
Keywords
- Actuators (1)
- Agenda 2030 (1)
- Anisotropic hyperelasticity (1)
- Antenna arrays (1)
- Application Integration (1)
- Automotive Industry (1)
- Basic prices (1)
- Bernstein Basis (1)
- Bernstein polynomial (2)
- Bio-vital data (1)
Institute
- Fakultät Architektur und Gestaltung (1)
- Fakultät Bauingenieurwesen (1)
- Fakultät Informatik (15)
- Fakultät Maschinenbau (3)
- Fakultät Wirtschafts-, Kultur- und Rechtswissenschaften (5)
- Institut für Angewandte Forschung - IAF (5)
- Institut für Optische Systeme - IOS (13)
- Institut für Strategische Innovation und Technologiemanagement - IST (4)
- Institut für Systemdynamik - ISD (14)
- Konstanz Institut für Corporate Governance - KICG (1)
Deep neural networks have become a veritable alternative to classic speaker recognition and clustering methods in recent years. However, while the speech signal clearly is a time series, and despite the body of literature on the benefits of prosodic (suprasegmental) features, identifying voices has usually not been approached with sequence learning methods. Only recently has a recurrent neural network (RNN) been successfully applied to this task, while the use of convolutional neural networks (CNNs) (that are not able to capture arbitrary time dependencies, unlike RNNs) still prevails. In this paper, we show the effectiveness of RNNs for speaker recognition by improving state of the art speaker clustering performance and robustness on the classic TIMIT benchmark. We provide arguments why RNNs are superior by experimentally showing a “sweet spot” of the segment length for successfully capturing prosodic information that has been theoretically predicted in previous work.
Research on Shadow IT is facing a conceptual dilemma in cases where previously "covert" systems developed by business entities (individual users, business workgroups, or business units) are integrated in the organizational IT management. These systems become visible, are therefore not "in the shadows" anymore, and subsequently do not fit to existing definitions of Shadow IT. Practice shows that some information systems share characteristics of Shadow IT, but are created openly in alignment with the IT department. This paper therefore proposes the term "Business-managed IT" to describe "overt" information systems developed or managed by business entities. We distinguish Business-managed IT from Shadow IT by illustrating case vignettes. Accordingly, our contribution is to suggest a concept and its delineation against other concepts. In this way, IS researchers interested in IT originated from or maintained by business entities can construct theories with a wider scope of application that are at the same time more specific to practical problems. In addition, value-laden terminology is complemented by a vocabulary that values potentially innovative developments by business entities more adequately. From a practical point of view, the distinction can be used to discuss the distribution of task responsibilities for information systems.
This thesis considers bounding functions for multivariate polynomials and rational functions over boxes and simplices. It also considers the synthesis of polynomial Lyapunov functions for obtaining the stability of control systems. Bounding the range of functions is an important issue in many areas of mathematics and its applications like global optimization, computer aided geometric design, robust control etc.
Objective: This paper presents an algorithm for non-invasive sleep stage identification using respiratory, heart rate and movement signals. The algorithm is part of a system suitable for long-term monitoring in a home environment, which should support experts analysing sleep. Approach: As there is a strong correlation between bio-vital signals and sleep stages, multinomial logistic regression was chosen for categorical distribution of sleep stages. Several derived parameters of three signals (respiratory, heart rate and movement) are input for the proposed method. Sleep recordings of five subjects were used for the training of a machine learning model and 30 overnight recordings collected from 30 individuals with about 27 000 epochs of 30 s intervals each were evaluated. Main results: The achieved rate of accuracy is 72% for Wake, NREM, REM (with Cohen's kappa value 0.67) and 58% for Wake, Light (N1 and N2), Deep (N3) and REM stages (Cohen's kappa is 0.50). Our approach has confirmed the potential of this method and disclosed several ways for its improvement. Significance: The results indicate that respiratory, heart rate and movement signals can be used for sleep studies with a reasonable level of accuracy. These inputs can be obtained in a non-invasive way applying it in a home environment. The proposed system introduces a convenient approach for a long-term monitoring system which could support sleep laboratories. The algorithm which was developed allows for an easy adjustment of input parameters that depend on available signals and for this reason could also be used with various hardware systems.
Input–Output modellers are often faced with the task of estimating missing Use tables at basic prices and also valuation matrices of the individual countries. This paper examines a selection of estimation methods applied to the European context where the analysts are not in possession of superior data. The estimation methods are restricted to the use of automated methods that would require more than just the row and column sums of the tables (as in projections) but less than a combination of various conflicting information (as in compilation). The results are assessed against the official Supply, Use and Input–Output tables of Belgium, Germany, Italy, Netherlands, Finland, Austria and Slovakia by using matrix difference metrics. The main conclusion is that using the structures of previous years usually performs better than any other approach.
This paper describes the effectiveness and efficiency of Virtual Reality training during a commissioning process. Therefore, 500 picking orders with more than 2000 part-picking operations with 30 test persons have been conducted and analyzed in the Modellfabrik Bodensee. The study points out the advantages and disadvantages of virtual training in comparison to a real execution of a picking process with and without any training.
The paper investigates an innovative actuator combination based on the magnetic shape memory technology. The actuator is composed of an electromagnet, which is activated to produce motion, and a magnetic shape memory element, which is used passively to yield multistability, i.e. the possibility of holding a position without input power. Based on the experimental open-loop frequency characterization of the actuator, a position controller is developed and tested in several experiments.