004 Datenverarbeitung; Informatik
Refine
Has Fulltext
- yes (78)
Is part of the Bibliography
- yes (78)
Year of publication
Document Type
- Journal article (78) (remove)
Keywords
- virtual reality (11)
- machine learning (5)
- augmented reality (3)
- immersion (3)
- Deep learning (2)
- Quadrocopter (2)
- Quadrotor (2)
- XR (2)
- artificial intelligence (2)
- automation (2)
Institute
- Institut für Informatik (78) (remove)
In many real world settings, imbalanced data impedes model performance of learning algorithms, like neural networks, mostly for rare cases. This is especially problematic for tasks focusing on these rare occurrences. For example, when estimating precipitation, extreme rainfall events are scarce but important considering their potential consequences. While there are numerous well studied solutions for classification settings, most of them cannot be applied to regression easily. Of the few solutions for regression tasks, barely any have explored cost-sensitive learning which is known to have advantages compared to sampling-based methods in classification tasks. In this work, we propose a sample weighting approach for imbalanced regression datasets called DenseWeight and a cost-sensitive learning approach for neural network regression with imbalanced data called DenseLoss based on our weighting scheme. DenseWeight weights data points according to their target value rarities through kernel density estimation (KDE). DenseLoss adjusts each data point’s influence on the loss according to DenseWeight, giving rare data points more influence on model training compared to common data points. We show on multiple differently distributed datasets that DenseLoss significantly improves model performance for rare data points through its density-based weighting scheme. Additionally, we compare DenseLoss to the state-of-the-art method SMOGN, finding that our method mostly yields better performance. Our approach provides more control over model training as it enables us to actively decide on the trade-off between focusing on common or rare cases through a single hyperparameter, allowing the training of better models for rare data points.
To deliver the best user experience (UX), the human-centered design cycle (HCDC) serves as a well-established guideline to application developers. However, it does not yet cover network-specific requirements, which become increasingly crucial, as most applications deliver experience over the Internet. The missing network-centric view is provided by Quality of Experience (QoE), which could team up with UX towards an improved overall experience. By considering QoE aspects during the development process, it can be achieved that applications become network-aware by design. In this paper, the Quality of Experience Centered Design Cycle (QoE-CDC) is proposed, which provides guidelines on how to design applications with respect to network-specific requirements and QoE. Its practical value is showcased for popular application types and validated by outlining the design of a new smartphone application. We show that combining HCDC and QoE-CDC will result in an application design, which reaches a high UX and avoids QoE degradation.
In recent history, normalized digital surface models (nDSMs) have been constantly gaining importance as a means to solve large-scale geographic problems. High-resolution surface models are precious, as they can provide detailed information for a specific area. However, measurements with a high resolution are time consuming and costly. Only a few approaches exist to create high-resolution nDSMs for extensive areas. This article explores approaches to extract high-resolution nDSMs from low-resolution Sentinel-2 data, allowing us to derive large-scale models. We thereby utilize the advantages of Sentinel 2 being open access, having global coverage, and providing steady updates through a high repetition rate. Several deep learning models are trained to overcome the gap in producing high-resolution surface maps from low-resolution input data. With U-Net as a base architecture, we extend the capabilities of our model by integrating tailored multiscale encoders with differently sized kernels in the convolution as well as conformed self-attention inside the skip connection gates. Using pixelwise regression, our U-Net base models can achieve a mean height error of approximately 2 m. Moreover, through our enhancements to the model architecture, we reduce the model error by more than 7%.
An innovative framework has been developed for teamwork of two quadcopter formations, each having its specified formation geometry, assigned task, and matching control scheme. Position control for quadcopters in one of the formations has been implemented through a Linear Quadratic Regulator Proportional Integral (LQR PI) control scheme based on explicit model following scheme. Quadcopters in the other formation are controlled through LQR PI servomechanism control scheme. These two control schemes are compared in terms of their performance and control effort. Both formations are commanded by respective ground stations through virtual leaders. Quadcopters in formations are able to track desired trajectories as well as hovering at desired points for selected time duration. In case of communication loss between ground station and any of the quadcopters, the neighboring quadcopter provides the command data, received from the ground station, to the affected unit. Proposed control schemes have been validated through extensive simulations using MATLAB®/Simulink® that provided favorable results.
Mapping and localization of mobile robots in an unknown environment are essential for most high-level operations like autonomous navigation or exploration. This paper presents a novel approach for combining estimated trajectories, namely curvefusion. The robot used in the experiments is equipped with a horizontally mounted 2D profiler, a constantly spinning 3D laser scanner and a GPS module. The proposed algorithm first combines trajectories from different sensors to optimize poses of the planar three degrees of freedom (DoF) trajectory, which is then fed into continuous-time simultaneous localization and mapping (SLAM) to further improve the trajectory. While state-of-the-art multi-sensor fusion methods mainly focus on probabilistic methods, our approach instead adopts a deformation-based method to optimize poses. To this end, a similarity metric for curved shapes is introduced into the robotics community to fuse the estimated trajectories. Additionally, a shape-based point correspondence estimation method is applied to the multi-sensor time calibration. Experiments show that the proposed fusion method can achieve relatively better accuracy, even if the error of the trajectory before fusion is large, which demonstrates that our method can still maintain a certain degree of accuracy in an environment where typical pose estimation methods have poor performance. In addition, the proposed time-calibration method also achieves high accuracy in estimating point correspondences.
Cover contact graphs
(2012)
We study problems that arise in the context of covering certain geometric objects called seeds (e.g., points or disks) by a set of other geometric objects called cover (e.g., a set of disks or homothetic triangles). We insist that the interiors of the seeds and the cover elements are pairwise disjoint, respectively, but they can touch. We call the contact graph of a cover a cover contact graph (CCG). We are interested in three types of tasks, both in the general case and in the special case of seeds on a line: (a) deciding whether a given seed set has a connected CCG, (b) deciding whether a given graph has a realization as a CCG on a given seed set, and (c) bounding the sizes of certain classes of CCG’s. Concerning (a) we give efficient algorithms for the case that seeds are points and show that the problem becomes hard if seeds and covers are disks. Concerning (b) we show that this problem is hard even for point seeds and disk covers (given a fixed correspondence between graph vertices and seeds). Concerning (c) we obtain upper and lower bounds on the number of CCG’s for point seeds.
Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, particularly systematic and location-specific representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we assess the use of convolutional Deep Learning climate MOS approaches and present the ConvMOS architecture which is specifically designed based on the observation that there are systematic and location-specific errors in the precipitation estimates of climate models. We apply ConvMOS models to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters for reducing location-specific errors and global model parameters for reducing systematic errors is indeed beneficial for MOS performance. We find that ConvMOS models can reduce errors considerably and perform significantly better than three commonly used MOS approaches and plain ResNet and U-Net models in most cases. Our results show that non-linear MOS models underestimate the number of extreme precipitation events, which we alleviate by training models specialized towards extreme precipitation events with the imbalanced regression method DenseLoss. While we consider climate MOS, we argue that aspects of ConvMOS may also be beneficial in other domains with geospatial data, such as air pollution modeling or weather forecasts.
This article presents a novel method for controlling a virtual audience system (VAS) in Virtual Reality (VR) application, called STAGE, which has been originally designed for supervised public speaking training in university seminars dedicated to the preparation and delivery of scientific talks. We are interested in creating pedagogical narratives: narratives encompass affective phenomenon and rather than organizing events changing the course of a training scenario, pedagogical plans using our system focus on organizing the affects it arouses for the trainees. Efficiently controlling a virtual audience towards a specific training objective while evaluating the speaker’s performance presents a challenge for a seminar instructor: the high level of cognitive and physical demands required to be able to control the virtual audience, whilst evaluating speaker’s performance, adjusting and allowing it to quickly react to the user’s behaviors and interactions. It is indeed a critical limitation of a number of existing systems that they rely on a Wizard of Oz approach, where the tutor drives the audience in reaction to the user’s performance. We address this problem by integrating with a VAS a high-level control component for tutors, which allows using predefined audience behavior rules, defining custom ones, as well as intervening during run-time for finer control of the unfolding of the pedagogical plan. At its core, this component offers a tool to program, select, modify and monitor interactive training narratives using a high-level representation. The STAGE offers the following features: i) a high-level API to program pedagogical narratives focusing on a specific public speaking situation and training objectives, ii) an interactive visualization interface iii) computation and visualization of user metrics, iv) a semi-autonomous virtual audience composed of virtual spectators with automatic reactions to the speaker and surrounding spectators while following the pedagogical plan V) and the possibility for the instructor to embody a virtual spectator to ask questions or guide the speaker from within the Virtual Environment. We present here the design, and implementation of the tutoring system and its integration in STAGE, and discuss its reception by end-users.
Presence is often considered the most important quale describing the subjective feeling of being in a computer-generated and/or computer-mediated virtual environment. The identification and separation of orthogonal presence components, i.e., the place illusion and the plausibility illusion, has been an accepted theoretical model describing Virtual Reality (VR) experiences for some time. This perspective article challenges this presence-oriented VR theory. First, we argue that a place illusion cannot be the major construct to describe the much wider scope of virtual, augmented, and mixed reality (VR, AR, MR: or XR for short). Second, we argue that there is no plausibility illusion but merely plausibility, and we derive the place illusion caused by the congruent and plausible generation of spatial cues and similarly for all the current model’s so-defined illusions. Finally, we propose congruence and plausibility to become the central essential conditions in a novel theoretical model describing XR experiences and effects.
Scalability is often mentioned in literature, but a stringent definition is missing. In particular, there is no general scalability assessment which clearly indicates whether a system scales or not or whether a system scales better than another. The key contribution of this article is the definition of a scalability index (SI) which quantifies if a system scales in comparison to another system, a hypothetical system, e.g., linear system, or the theoretically optimal system. The suggested SI generalizes different metrics from literature, which are specialized cases of our SI. The primary target of our scalability framework is, however, benchmarking of two systems, which does not require any reference system. The SI is demonstrated and evaluated for different use cases, that are (1) the performance of an IoT load balancer depending on the system load, (2) the availability of a communication system depending on the size and structure of the network, (3) scalability comparison of different location selection mechanisms in fog computing with respect to delays and energy consumption; (4) comparison of time-sensitive networking (TSN) mechanisms in terms of efficiency and utilization. Finally, we discuss how to use and how not to use the SI and give recommendations and guidelines in practice. To the best of our knowledge, this is the first work which provides a general SI for the comparison and benchmarking of systems, which is the primary target of our scalability analysis.
CLIP knows image aesthetics
(2022)
Most Image Aesthetic Assessment (IAA) methods use a pretrained ImageNet classification model as a base to fine-tune. We hypothesize that content classification is not an optimal pretraining task for IAA, since the task discourages the extraction of features that are useful for IAA, e.g., composition, lighting, or style. On the other hand, we argue that the Contrastive Language-Image Pretraining (CLIP) model is a better base for IAA models, since it has been trained using natural language supervision. Due to the rich nature of language, CLIP needs to learn a broad range of image features that correlate with sentences describing the image content, composition, environments, and even subjective feelings about the image. While it has been shown that CLIP extracts features useful for content classification tasks, its suitability for tasks that require the extraction of style-based features like IAA has not yet been shown. We test our hypothesis by conducting a three-step study, investigating the usefulness of features extracted by CLIP compared to features obtained from the last layer of a comparable ImageNet classification model. In each step, we get more computationally expensive. First, we engineer natural language prompts that let CLIP assess an image's aesthetic without adjusting any weights in the model. To overcome the challenge that CLIP's prompting only is applicable to classification tasks, we propose a simple but effective strategy to convert multiple prompts to a continuous scalar as required when predicting an image's mean aesthetic score. Second, we train a linear regression on the AVA dataset using image features obtained by CLIP's image encoder. The resulting model outperforms a linear regression trained on features from an ImageNet classification model. It also shows competitive performance with fully fine-tuned networks based on ImageNet, while only training a single layer. Finally, by fine-tuning CLIP's image encoder on the AVA dataset, we show that CLIP only needs a fraction of training epochs to converge, while also performing better than a fine-tuned ImageNet model. Overall, our experiments suggest that CLIP is better suited as a base model for IAA methods than ImageNet pretrained networks.
Mindfulness is considered an important factor of an individual's subjective well-being. Consequently, Human-Computer Interaction (HCI) has investigated approaches that strengthen mindfulness, i.e., by inventing multimedia technologies to support mindfulness meditation. These approaches often use smartphones, tablets, or consumer-grade desktop systems to allow everyday usage in users' private lives or in the scope of organized therapies. Virtual, Augmented, and Mixed Reality (VR, AR, MR; in short: XR) significantly extend the design space for such approaches. XR covers a wide range of potential sensory stimulation, perceptive and cognitive manipulations, content presentation, interaction, and agency. These facilities are linked to typical XR-specific perceptions that are conceptually closely related to mindfulness research, such as (virtual) presence and (virtual) embodiment. However, a successful exploitation of XR that strengthens mindfulness requires a systematic analysis of the potential interrelation and influencing mechanisms between XR technology, its properties, factors, and phenomena and existing models and theories of the construct of mindfulness. This article reports such a systematic analysis of XR-related research from HCI and life sciences to determine the extent to which existing research frameworks on HCI and mindfulness can be applied to XR technologies, the potential of XR technologies to support mindfulness, and open research gaps. Fifty papers of ACM Digital Library and National Institutes of Health's National Library of Medicine (PubMed) with and without empirical efficacy evaluation were included in our analysis. The results reveal that at the current time, empirical research on XR-based mindfulness support mainly focuses on therapy and therapeutic outcomes. Furthermore, most of the currently investigated XR-supported mindfulness interactions are limited to vocally guided meditations within nature-inspired virtual environments. While an analysis of empirical research on those systems did not reveal differences in mindfulness compared to non-mediated mindfulness practices, various design proposals illustrate that XR has the potential to provide interactive and body-based innovations for mindfulness practice. We propose a structured approach for future work to specify and further explore the potential of XR as mindfulness-support. The resulting framework provides design guidelines for XR-based mindfulness support based on the elements and psychological mechanisms of XR interactions.
This article presents an immersive virtual reality (VR) system for training classroom management skills, with a specific focus on learning to manage disruptive student behavior in face-to-face, one-to-many teaching scenarios. The core of the system is a real-time 3D virtual simulation of a classroom populated by twenty-four semi-autonomous virtual students. The system has been designed as a companion tool for classroom management seminars in a syllabus for primary and secondary school teachers. This will allow lecturers to link theory with practice using the medium of VR. The system is therefore designed for two users: a trainee teacher and an instructor supervising the training session. The teacher is immersed in a real-time 3D simulation of a classroom by means of a head-mounted display and headphone. The instructor operates a graphical desktop console, which renders a view of the class and the teacher whose avatar movements are captured by a marker less tracking system. This console includes a 2D graphics menu with convenient behavior and feedback control mechanisms to provide human-guided training sessions. The system is built using low-cost consumer hardware and software. Its architecture and technical design are described in detail. A first evaluation confirms its conformance to critical usability requirements (i.e., safety and comfort, believability, simplicity, acceptability, extensibility, affordability, and mobility). Our initial results are promising and constitute the necessary first step toward a possible investigation of the efficiency and effectiveness of such a system in terms of learning outcomes and experience.
This paper describes the estimation of the body weight of a person in front of an RGB-D camera. A survey of different methods for body weight estimation based on depth sensors is given. First, an estimation of people standing in front of a camera is presented. Second, an approach based on a stream of depth images is used to obtain the body weight of a person walking towards a sensor. The algorithm first extracts features from a point cloud and forwards them to an artificial neural network (ANN) to obtain an estimation of body weight. Besides the algorithm for the estimation, this paper further presents an open-access dataset based on measurements from a trauma room in a hospital as well as data from visitors of a public event. In total, the dataset contains 439 measurements. The article illustrates the efficiency of the approach with experiments with persons lying down in a hospital, standing persons, and walking persons. Applicable scenarios for the presented algorithm are body weight-related dosing of emergency patients.
The development of ICT infrastructures has facilitated the emergence of new paradigms for looking at society and the environment over the last few years. Participatory environmental sensing, i.e. directly involving citizens in environmental monitoring, is one example, which is hoped to encourage learning and enhance awareness of environmental issues. In this paper, an analysis of the behaviour of individuals involved in noise sensing is presented. Citizens have been involved in noise measuring activities through the WideNoise smartphone application. This application has been designed to record both objective (noise samples) and subjective (opinions, feelings) data. The application has been open to be used freely by anyone and has been widely employed worldwide. In addition, several test cases have been organised in European countries. Based on the information submitted by users, an analysis of emerging awareness and learning is performed. The data show that changes in the way the environment is perceived after repeated usage of the application do appear. Specifically, users learn how to recognise different noise levels they are exposed to. Additionally, the subjective data collected indicate an increased user involvement in time and a categorisation effect between pleasant and less pleasant environments.
Background
Colorectal cancer is a leading cause of cancer-related deaths worldwide. The best method to prevent CRC is a colonoscopy. However, not all colon polyps have the risk of becoming cancerous. Therefore, polyps are classified using different classification systems. After the classification, further treatment and procedures are based on the classification of the polyp. Nevertheless, classification is not easy. Therefore, we suggest two novel automated classifications system assisting gastroenterologists in classifying polyps based on the NICE and Paris classification.
Methods
We build two classification systems. One is classifying polyps based on their shape (Paris). The other classifies polyps based on their texture and surface patterns (NICE). A two-step process for the Paris classification is introduced: First, detecting and cropping the polyp on the image, and secondly, classifying the polyp based on the cropped area with a transformer network. For the NICE classification, we design a few-shot learning algorithm based on the Deep Metric Learning approach. The algorithm creates an embedding space for polyps, which allows classification from a few examples to account for the data scarcity of NICE annotated images in our database.
Results
For the Paris classification, we achieve an accuracy of 89.35 %, surpassing all papers in the literature and establishing a new state-of-the-art and baseline accuracy for other publications on a public data set. For the NICE classification, we achieve a competitive accuracy of 81.13 % and demonstrate thereby the viability of the few-shot learning paradigm in polyp classification in data-scarce environments. Additionally, we show different ablations of the algorithms. Finally, we further elaborate on the explainability of the system by showing heat maps of the neural network explaining neural activations.
Conclusion
Overall we introduce two polyp classification systems to assist gastroenterologists. We achieve state-of-the-art performance in the Paris classification and demonstrate the viability of the few-shot learning paradigm in the NICE classification, addressing the prevalent data scarcity issues faced in medical machine learning.
In this paper we study connectivity augmentation problems. Given a connected graph G with some desirable property, we want to make G 2-vertex connected (or 2-edge connected) by adding edges such that the resulting graph keeps the property. The aim is to add as few edges as possible. The property that we consider is planarity, both in an abstract graph-theoretic and in a geometric setting, where vertices correspond to points in the plane and edges to straight-line segments.
We show that it is NP-hard to nd a minimum-cardinality augmentation that makes a planar graph 2-edge connected. For making a planar graph 2-vertex connected this was known. We further show that both problems are hard in the geometric setting, even when restricted to trees. The problems remain hard for higher degrees of connectivity. On the other hand we give polynomial-time algorithms for the special case of convex geometric graphs.
We also study the following related problem. Given a planar (plane geometric) graph G, two vertices s and t of G, and an integer c, how many edges have to be added to G such that G is still planar (plane geometric) and contains c edge- (or vertex-) disjoint s{t paths? For the planar case we give a linear-time algorithm for c = 2. For the plane geometric case we give optimal worst-case bounds for c = 2; for c = 3 we characterize the cases that have a solution.
Einleitung:
Multiple-Choice-Klausuren spielen immer noch eine herausragende Rolle für fakultätsinterne medizinische Prüfungen. Neben inhaltlichen Arbeiten stellt sich die Frage, wie die technische Abwicklung optimiert werden kann. Für Dozenten in der Medizin gibt es zunehmend drei Optionen zur Durchführung von MC-Klausuren: Papierklausuren mit oder ohne Computerunterstützung oder vollständig elektronische Klausuren. Kritische Faktoren sind der Aufwand für die Formatierung der Klausur, der logistische Aufwand bei der Klausurdurchführung, die Qualität, Schnelligkeit und der Aufwand der Klausurkorrektur, die Bereitstellung der Dokumente für die Einsichtnahme, und die statistische Analyse der Klausurergebnisse.
Methoden:
An der Universität Würzburg wird seit drei Semestern ein Computerprogramm zur Eingabe und Formatierung der MC-Fragen in medizinischen und anderen Papierklausuren verwendet und optimiert, mit dem im Wintersemester (WS) 2009/2010 elf, im Sommersemester (SS) 2010 zwölf und im WS 2010/11 dreizehn medizinische Klausuren erstellt und anschließend die eingescannten Antwortblätter automatisch ausgewertet wurden. In den letzten beiden Semestern wurden die Aufwände protokolliert.
Ergebnisse:
Der Aufwand der Formatierung und der Auswertung einschl. nachträglicher Anpassung der Auswertung einer Durchschnittsklausur mit ca. 140 Teilnehmern und ca. 35 Fragen ist von 5-7 Stunden für Klausuren ohne Komplikation im WS 2009/2010 über ca. 2 Stunden im SS 2010 auf ca. 1,5 Stunden im WS 2010/11 gefallen. Einschließlich der Klausuren mit Komplikationen bei der Auswertung betrug die durchschnittliche Zeit im SS 2010 ca. 3 Stunden und im WS 10/11 ca. 2,67 Stunden pro Klausur.
Diskussion:
Für konventionelle Multiple-Choice-Klausuren bietet die computergestützte Formatierung und Auswertung von Papierklausuren einen beträchtlichen Zeitvorteil für die Dozenten im Vergleich zur manuellen Korrektur von Papierklausuren und benötigt im Vergleich zu rein elektronischen Klausuren eine deutlich einfachere technische Infrastruktur und weniger Personal bei der Klausurdurchführung.
Knowledge about ransomware is important for protecting sensitive data and for participating in public debates about suitable regulation regarding its security. However, as of now, this topic has received little to no attention in most school curricula. As such, it is desirable to analyze what citizens can learn about this topic outside of formal education, e.g., from news articles. This analysis is both relevant to analyzing the public discourse about ransomware, as well as to identify what aspects of this topic should be included in the limited time available for this topic in formal education. Thus, this paper was motivated both by educational and media research. The central goal is to explore how the media reports on this topic and, additionally, to identify potential misconceptions that could stem from this reporting. To do so, we conducted an exploratory case study into the reporting of 109 media articles regarding a high-impact ransomware event: the shutdown of the Colonial Pipeline (located in the east of the USA). We analyzed how the articles introduced central terminology, what details were provided, what details were not, and what (mis-)conceptions readers might receive from them. Our results show that an introduction of the terminology and technical concepts of security is insufficient for a complete understanding of the incident. Most importantly, the articles may lead to four misconceptions about ransomware that are likely to lead to misleading conclusions about the responsibility for the incident and possible political and technical options to prevent such attacks in the future.