Refine
Year of publication
Institute
- Fachbereich Medizintechnik und Technomathematik (241) (remove)
Document Type
- Conference Proceeding (241) (remove)
Keywords
- Biosensor (25)
- CAD (11)
- Finite-Elemente-Methode (11)
- civil engineering (11)
- Bauingenieurwesen (10)
- Einspielen <Werkstoff> (6)
- shakedown analysis (6)
- Clusterion (4)
- Limit analysis (4)
- Natural language processing (4)
Reliable methods for automatic readability assessment have the potential to impact a variety of fields, ranging from machine translation to self-informed learning. Recently, large language models for the German language (such as GBERT and GPT-2-Wechsel) have become available, allowing to develop Deep Learning based approaches that promise to further improve automatic readability assessment. In this contribution, we studied the ability of ensembles of fine-tuned GBERT and GPT-2-Wechsel models to reliably predict the readability of German sentences. We combined these models with linguistic features and investigated the dependence of prediction performance on ensemble size and composition. Mixed ensembles of GBERT and GPT-2-Wechsel performed better than ensembles of the same size consisting of only GBERT or GPT-2-Wechsel models. Our models were evaluated in the GermEval 2022 Shared Task on Text Complexity Assessment on data of German sentences. On out-of-sample data, our best ensemble achieved a root mean squared error of 0:435.
Messenger apps like WhatsApp or Telegram are an integral part of daily communication. Besides the various positive effects, those services extend the operating range of criminals. Open trading groups with many thousand participants emerged on Telegram. Law enforcement agencies monitor suspicious users in such chat rooms. This research shows that text analysis, based on natural language processing, facilitates this through a meaningful domain overview and detailed investigations. We crawled a corpus from such self-proclaimed black markets and annotated five attribute types products, money, payment methods, user names, and locations. Based on each message a user sends, we extract and group these attributes to build profiles. Then, we build features to cluster the profiles. Pretrained word vectors yield better unsupervised clustering results than current
state-of-the-art transformer models. The result is a semantically meaningful high-level overview of the user landscape of black market chatrooms. Additionally, the extracted structured information serves as a foundation for further data exploration, for example, the most active users or preferred payment methods.
Conventional EEG devices cannot be used in everyday life and hence, past decade research has been focused on Ear-EEG for mobile, at-home monitoring for various applications ranging from emotion detection to sleep monitoring. As the area available for electrode contact in the ear is limited, the electrode size and location play a vital role for an Ear-EEG system. In this investigation, we present a quantitative study of ear-electrodes with two electrode sizes at different locations in a wet and dry configuration. Electrode impedance scales inversely with size and ranges from 450 kΩ to 1.29 MΩ for dry and from 22 kΩ to 42 kΩ for wet contact at 10 Hz. For any size, the location in the ear canal with the lowest impedance is ELE (Left Ear Superior), presumably due to increased contact pressure caused by the outer-ear anatomy. The results can be used to optimize signal pickup and SNR for specific applications. We demonstrate this by recording sleep spindles during sleep onset with high quality (5.27 μVrms).
Multi-attribute relation extraction (MARE): simplifying the application of relation extraction
(2021)
Natural language understanding’s relation extraction makes innovative and encouraging novel business concepts possible and facilitates new digitilized decision-making processes. Current approaches allow the extraction of relations with a fixed number of entities as attributes. Extracting relations with an arbitrary amount of attributes requires complex systems and costly relation-trigger annotations to assist these systems. We introduce multi-attribute relation extraction (MARE) as an assumption-less problem formulation with two approaches, facilitating an explicit mapping from business use cases to the data annotations. Avoiding elaborated annotation constraints simplifies the application of relation extraction approaches. The evaluation compares our models to current state-of-the-art event extraction and binary relation extraction methods. Our approaches show improvement compared to these on the extraction of general multi-attribute relations.
The integration of frequently changing, volatile product data from different manufacturers into a single catalog is a significant challenge for small and medium-sized e-commerce companies. They rely on timely integrating product data to present them aggregated in an online shop without knowing format specifications, concept understanding of manufacturers, and data quality. Furthermore, format, concepts, and data quality may change at any time. Consequently, integrating product catalogs into a single standardized catalog is often a laborious manual task. Current strategies to streamline or automate catalog integration use techniques based on machine learning, word vectorization, or semantic similarity. However, most approaches struggle with low-quality or real-world data. We propose Attribute Label Ranking (ALR) as a recommendation engine to simplify the integration process of previously unknown, proprietary tabular format into a standardized catalog for practitioners. We evaluate ALR by focusing on the impact of different neural network architectures, language features, and semantic similarity. Additionally, we consider metrics for industrial application and present the impact of ALR in production and its limitations.
The progress in natural language processing (NLP) research over the last years, offers novel business opportunities for companies, as automated user interaction or improved data analysis. Building sophisticated NLP applications requires dealing with modern machine learning (ML) technologies, which impedes enterprises from establishing successful NLP projects. Our experience in applied NLP research projects shows that the continuous integration of research prototypes in production-like environments with quality assurance builds trust in the software and shows convenience and usefulness regarding the business goal. We introduce STAMP 4 NLP as an iterative and incremental process model for developing NLP applications. With STAMP 4 NLP, we merge software engineering principles with best practices from data science. Instantiating our process model allows efficiently creating prototypes by utilizing templates, conventions, and implementations, enabling developers and data scientists to focus on the business goals. Due to our iterative-incremental approach, businesses can deploy an enhanced version of the prototype to their software environment after every iteration, maximizing potential business value and trust early and avoiding the cost of successful yet never deployed experiments.
In positron emission tomography improving time, energy and spatial detector resolutions and using Compton kinematics introduces the possibility to reconstruct a radioactivity distribution image from scatter coincidences, thereby enhancing image quality. The number of single scattered coincidences alone is in the same order of magnitude as true coincidences. In this work, a compact Compton camera module based on monolithic scintillation material is investigated as a detector ring module. The detector interactions are simulated with Monte Carlo package GATE. The scattering angle inside the tissue is derived from the energy of the scattered photon, which results in a set of possible scattering trajectories or broken line of response. The Compton kinematics collimation reduces the number of solutions. Additionally, the time of flight information helps localize the position of the annihilation. One of the questions of this investigation is related to how the energy, spatial and temporal resolutions help confine the possible annihilation volume. A comparison of currently technically feasible detector resolutions (under laboratory conditions) demonstrates the influence on this annihilation volume and shows that energy and coincidence time resolution have a significant impact. An enhancement of the latter from 400 ps to 100 ps leads to a smaller annihilation volume of around 50%, while a change of the energy resolution in the absorber layer from 12% to 4.5% results in a reduction of 60%. The inclusion of single tissue-scattered data has the potential to increase the sensitivity of a scanner by a factor of 2 to 3 times. The concept can be further optimized and extended for multiple scatter coincidences and subsequently validated by a reconstruction algorithm.
A new formulation to calculate the shakedown limit load of Kirchhoff plates under stochastic conditions of strength is developed. Direct structural reliability design by chance con-strained programming is based on the prescribed failure probabilities, which is an effective approach of stochastic programming if it can be formulated as an equivalent deterministic optimization problem. We restrict uncertainty to strength, the loading is still deterministic. A new formulation is derived in case of random strength with lognormal distribution. Upper bound and lower bound shakedown load factors are calculated simultaneously by a dual algorithm.
The paper presents a method for the quantitative assessment of choroidal blood flow using an OCT-A system. The developed technique for processing of OCT-A scans is divided into two stages. At the first stage, the identification of the boundaries in the selected portion was performed. At the second stage, each pixel mark on the selected layer was represented as a volume unit, a voxel, which characterizes the region of moving blood. Three geometric shapes were considered to represent the voxel. On the example of one OCT-A scan, this work presents a quantitative assessment of the blood flow index. A possible modification of two-stage algorithm based on voxel scan processing is presented.