004 Datenverarbeitung; Informatik
Refine
Has Fulltext
- yes (285)
Year of publication
Document Type
- Journal article (127)
- Doctoral Thesis (80)
- Working Paper (37)
- Preprint (19)
- Conference Proceeding (9)
- Jahresbericht (5)
- Master Thesis (4)
- Report (3)
- Other (1)
Language
- English (257)
- German (27)
- Multiple languages (1)
Keywords
- virtual reality (16)
- Datennetz (14)
- Leistungsbewertung (13)
- Quran (8)
- Robotik (8)
- Koran (7)
- Mobiler Roboter (7)
- Text Mining (7)
- Autonomer Roboter (6)
- Simulation (6)
Institute
- Institut für Informatik (203)
- Theodor-Boveri-Institut für Biowissenschaften (29)
- Institut Mensch - Computer - Medien (17)
- Institut für deutsche Philologie (17)
- Institut für Klinische Epidemiologie und Biometrie (7)
- Rechenzentrum (7)
- Center for Computational and Theoretical Biology (4)
- Graduate School of Science and Technology (3)
- Medizinische Klinik und Poliklinik II (3)
- Institut für Funktionsmaterialien und Biofabrikation (2)
Schriftenreihe
Sonstige beteiligte Institutionen
- Cologne Game Lab (2)
- Birmingham City University (1)
- DATE Lab, KITE Research Insititute, University Health Network, Toronto, Canada (1)
- EMBL Heidelberg (1)
- INAF Padova, Italy (1)
- Jacobs University Bremen, Germany (1)
- Open University of the Netherlands (1)
- Servicezentrum Medizin-Informatik (Universitätsklinikum) (1)
- Social and Technological Systems (SaTS) lab, School of Art, Media, Performance and Design, York University, Toronto, Canada (1)
- TH Köln (1)
The emerging serverless computing may meet Edge Cloud in a beneficial manner as the two offer flexibility and dynamicity in optimizing finite hardware resources. However, the lack of proper study of a joint platform leaves a gap in literature about consumption and performance of such integration. To this end, this paper identifies the key questions and proposes a methodology to answer them.
The Fifth Generation (5G) communication technology, its infrastructure and architecture, though already deployed in campus and small scale networks, is still undergoing continuous changes and research. Especially, in the light of future large scale deployments and industrial use cases, a detailed analysis of the performance and utilization with regard to latency and service times constraints is crucial. To this end, a fine granular investigation of the Network Function (NF) based core system and the duration for all the tasks performed by these services is necessary. This work presents the first steps towards analyzing the signaling traffic in 5G core networks, and introduces a tool to automatically extract sequence diagrams and service times for NF tasks from traffic traces.
Packets sent over a network can either get lost or reach their destination. Protocols like TCP try to solve this problem by resending the lost packets. However, retransmissions consume a lot of time and are cumbersome for the transmission of critical data. Multipath solutions are quite common to address this reliability issue and are available on almost every layer of the ISO/OSI model. We propose a solution based on a P4 network to duplicate packets in order to send them to their destination via multiple routes. The last network hop ensures that only a single copy of the traffic is further forwarded to its destination by adopting a concept similar to Bloom filters. Besides, if fast delivery is requested we provide a P4 prototype, which randomly forwards the packets over different transmission paths. For reproducibility, we implement our approach in a container-based network emulation system called Kathará.
Crowdsourced network measurements (CNMs) are becoming increasingly popular as they assess the performance of a mobile network from the end user's perspective on a large scale. Here, network measurements are performed directly on the end-users' devices, thus taking advantage of the real-world conditions end-users encounter. However, this type of uncontrolled measurement raises questions about its validity and reliability. The problem lies in the nature of this type of data collection. In CNMs, mobile network subscribers are involved to a large extent in the measurement process, and collect data themselves for the operator. The collection of data on user devices in arbitrary locations and at uncontrolled times requires means to ensure validity and reliability. To address this issue, our paper defines concepts and guidelines for analyzing the precision of CNMs; specifically, the number of measurements required to make valid statements. In addition to the formal definition of the aspect, we illustrate the problem and use an extensive sample data set to show possible assessment approaches. This data set consists of more than 20.4 million crowdsourced mobile measurements from across France, measured by a commercial data provider.
Given the growing interest of corporate stakeholders in Metaverse applications, there is a need to understand accessibility of these technologies for marginalized populations such as people living with dementia to ensure inclusive design of Metaverse applications. We assessed the accessibility of extended reality technology for people living with mild cognitive impairment and dementia to develop accessibility guidelines for these technologies. We used four strategies to synthesize evidence for barriers and facilitators of accessibility: (1) Findings from a non-systematic literature review, (2) guidelines from well-researched technology, (3) exploration of selected mixed reality technologies, and (4) observations from four sessions and video data of people living with dementia using mixed reality technologies. We utilized template analysis to develop codes and themes towards accessibility guidelines. Future work can validate our preliminary findings by applying them on video recordings or testing them in experiments.
Deep learning enables enormous progress in many computer vision-related tasks. Artificial Intel- ligence (AI) steadily yields new state-of-the-art results in the field of detection and classification. Thereby AI performance equals or exceeds human performance. Those achievements impacted many domains, including medical applications.
One particular field of medical applications is gastroenterology. In gastroenterology, machine learning algorithms are used to assist examiners during interventions. One of the most critical concerns for gastroenterologists is the development of Colorectal Cancer (CRC), which is one of the leading causes of cancer-related deaths worldwide. Detecting polyps in screening colonoscopies is the essential procedure to prevent CRC. Thereby, the gastroenterologist uses an endoscope to screen the whole colon to find polyps during a colonoscopy. Polyps are mucosal growths that can vary in severity.
This thesis supports gastroenterologists in their examinations with automated detection and clas- sification systems for polyps. The main contribution is a real-time polyp detection system. This system is ready to be installed in any gastroenterology practice worldwide using open-source soft- ware. The system achieves state-of-the-art detection results and is currently evaluated in a clinical trial in four different centers in Germany.
The thesis presents two additional key contributions: One is a polyp detection system with ex- tended vision tested in an animal trial. Polyps often hide behind folds or in uninvestigated areas. Therefore, the polyp detection system with extended vision uses an endoscope assisted by two additional cameras to see behind those folds. If a polyp is detected, the endoscopist receives a vi- sual signal. While the detection system handles the additional two camera inputs, the endoscopist focuses on the main camera as usual.
The second one are two polyp classification models, one for the classification based on shape (Paris) and the other on surface and texture (NBI International Colorectal Endoscopic (NICE) classification). Both classifications help the endoscopist with the treatment of and the decisions about the detected polyp.
The key algorithms of the thesis achieve state-of-the-art performance. Outstandingly, the polyp detection system tested on a highly demanding video data set shows an F1 score of 90.25 % while working in real-time. The results exceed all real-time systems in the literature. Furthermore, the first preliminary results of the clinical trial of the polyp detection system suggest a high Adenoma Detection Rate (ADR). In the preliminary study, all polyps were detected by the polyp detection system, and the system achieved a high usability score of 96.3 (max 100). The Paris classification model achieved an F1 score of 89.35 % which is state-of-the-art. The NICE classification model achieved an F1 score of 81.13 %.
Furthermore, a large data set for polyp detection and classification was created during this thesis. Therefore a fast and robust annotation system called Fast Colonoscopy Annotation Tool (FastCAT) was developed. The system simplifies the annotation process for gastroenterologists. Thereby the
i
gastroenterologists only annotate key parts of the endoscopic video. Afterward, those video parts are pre-labeled by a polyp detection AI to speed up the process. After the AI has pre-labeled the frames, non-experts correct and finish the annotation. This annotation process is fast and ensures high quality. FastCAT reduces the overall workload of the gastroenterologist on average by a factor of 20 compared to an open-source state-of-art annotation tool.
Within the healthcare environment, mobile health (mHealth) applications (apps) are becoming more and more important. The number of new mHealth apps has risen steadily in the last years. Especially the COVID-19 pandemic has led to an enormous amount of app releases. In most countries, mHealth applications have to be compliant with several regulatory aspects to be declared a “medical app”. However, the latest applicable medical device regulation (MDR) does not provide more details on the requirements for mHealth applications. When developing a medical app, it is essential that all contributors in an interdisciplinary team — especially software engineers — are aware of the specific regulatory requirements beforehand. The development process, however, should not be stalled due to integration of the MDR. Therefore, a developing framework that includes these aspects is required to facilitate a reliable and quick development process. The paper at hand introduces the creation of such a framework on the basis of the Corona Health and Corona Check apps. The relevant regulatory guidelines are listed and summarized as a guidance for medical app developments during the pandemic and beyond. In particular, the important stages and challenges faced that emerged during the entire development process are highlighted.
Visual stimuli are frequently used to improve memory, language learning or perception, and understanding of metacognitive processes. However, in virtual reality (VR), there are few systematically and empirically derived databases. This paper proposes the first collection of virtual objects based on empirical evaluation for inter-and transcultural encounters between English- and German-speaking learners. We used explicit and implicit measurement methods to identify cultural associations and the degree of stereotypical perception for each virtual stimuli (n = 293) through two online studies, including native German and English-speaking participants. The analysis resulted in a final well-describable database of 128 objects (called InteractionSuitcase). In future applications, the objects can be used as a great interaction or conversation asset and behavioral measurement tool in social VR applications, especially in the field of foreign language education. For example, encounters can use the objects to describe their culture, or teachers can intuitively assess stereotyped attitudes of the encounters.
Automatic image reconstruction is critical to cope with steadily increasing data from advanced microscopy. We describe here the Fiji macro 3D ART VeSElecT which we developed to study synaptic vesicles in electron tomograms. We apply this tool to quantify vesicle properties (i) in embryonic Danio rerio 4 and 8 days past fertilization (dpf) and (ii) to compare Caenorhabditis elegans N2 neuromuscular junctions (NMJ) wild-type and its septin mutant (unc-59(e261)). We demonstrate development-specific and mutant-specific changes in synaptic vesicle pools in both models. We confirm the functionality of our macro by applying our 3D ART VeSElecT on zebrafish NMJ showing smaller vesicles in 8 dpf embryos then 4 dpf, which was validated by manual reconstruction of the vesicle pool. Furthermore, we analyze the impact of C. elegans septin mutant unc-59(e261) on vesicle pool formation and vesicle size. Automated vesicle registration and characterization was implemented in Fiji as two macros (registration and measurement). This flexible arrangement allows in particular reducing false positives by an optional manual revision step. Preprocessing and contrast enhancement work on image-stacks of 1nm/pixel in x and y direction. Semi-automated cell selection was integrated. 3D ART VeSElecT removes interfering components, detects vesicles by 3D segmentation and calculates vesicle volume and diameter (spherical approximation, inner/outer diameter). Results are collected in color using the RoiManager plugin including the possibility of manual removal of non-matching confounder vesicles. Detailed evaluation considered performance (detected vesicles) and specificity (true vesicles) as well as precision and recall. We furthermore show gain in segmentation and morphological filtering compared to learning based methods and a large time gain compared to manual segmentation. 3D ART VeSElecT shows small error rates and its speed gain can be up to 68 times faster in comparison to manual annotation. Both automatic and semi-automatic modes are explained including a tutorial.
Interactive system for similarity-based inspection and assessment of the well-being of mHealth users
(2021)
Recent digitization technologies empower mHealth users to conveniently record their Ecological Momentary Assessments (EMA) through web applications, smartphones, and wearable devices. These recordings can help clinicians understand how the users' condition changes, but appropriate learning and visualization mechanisms are required for this purpose. We propose a web-based visual analytics tool, which processes clinical data as well as EMAs that were recorded through a mHealth application. The goals we pursue are (1) to predict the condition of the user in the near and the far future, while also identifying the clinical data that mostly contribute to EMA predictions, (2) to identify users with outlier EMA, and (3) to show to what extent the EMAs of a user are in line with or diverge from those users similar to him/her. We report our findings based on a pilot study on patient empowerment, involving tinnitus patients who recorded EMAs with the mHealth app TinnitusTips. To validate our method, we also derived synthetic data from the same pilot study. Based on this setting, results for different use cases are reported.
The holy grail of structural biology is to study a protein in situ, and this goal has been fast approaching since the resolution revolution and the achievement of atomic resolution. A cell's interior is not a dilute environment, and proteins have evolved to fold and function as needed in that environment; as such, an investigation of a cellular component should ideally include the full complexity of the cellular environment. Imaging whole cells in three dimensions using electron cryotomography is the best method to accomplish this goal, but it comes with a limitation on sample thickness and produces noisy data unamenable to direct analysis. This thesis establishes a novel workflow to systematically analyse whole-cell electron cryotomography data in three dimensions and to find and identify instances of protein complexes in the data to set up a determination of their structure and identity for success. Mycoplasma pneumoniae is a very small parasitic bacterium with fewer than 700 protein-coding genes, is thin enough and small enough to be imaged in large quantities by electron cryotomography, and can grow directly on the grids used for imaging, making it ideal for exploratory studies in structural proteomics. As part of the workflow, a methodology for training deep-learning-based particle-picking models is established.
As a proof of principle, a dataset of whole-cell Mycoplasma pneumoniae tomograms is used with this workflow to characterize a novel membrane-associated complex observed in the data. Ultimately, 25431 such particles are picked from 353 tomograms and refined to a density map with a resolution of 11 Å. Making good use of orthogonal datasets to filter search space and verify results, structures were predicted for candidate proteins and checked for suitable fit in the density map. In the end, with this approach, nine proteins were found to be part of the complex, which appears to be associated with chaperone activity and interact with translocon machinery.
Visual proteomics refers to the ultimate potential of in situ electron cryotomography: the comprehensive interpretation of tomograms. The workflow presented here is demonstrated to help in reaching that potential.
Heat and excessive solar radiation can produce abiotic stresses during apple maturation, resulting fruit quality. Therefore, the monitoring of temperature on fruit surface (FST) over the growing period can allow to identify thresholds, above of which several physiological disorders such as sunburn may occur in apple.
The current approaches neglect spatial variation of FST and have reduced repeatability, resulting in unreliable predictions. In this study, LiDAR laser scanning and thermal imaging were employed to detect the temperature on fruit surface by means of 3D point cloud. A process for calibrating the two sensors based on an active board target and producing a 3D thermal point cloud was suggested. After calibration, the sensor system was utilised to scan the fruit trees, while temperature values assigned in the corresponding 3D point cloud were based on the extrinsic calibration. Whereas a fruit detection algorithm was performed to segment the FST from each apple.
• The approach allows the calibration of LiDAR laser scanner with thermal camera in order to produce a 3D thermal point cloud.
• The method can be applied in apple trees for segmenting FST in 3D. Whereas the approach can be utilised to predict several physiological disorders including sunburn on fruit surface.
The strict restrictions introduced by the COVID-19 lockdowns, which started from March 2020, changed people’s daily lives and habits on many different levels. In this work, we investigate the impact of the lockdown on the communication behavior in the mobile instant messaging application WhatsApp. Our evaluations are based on a large dataset of 2577 private chat histories with 25,378,093 messages from 51,973 users. The analysis of the one-to-one and group conversations confirms that the lockdown severely altered the communication in WhatsApp chats compared to pre-pandemic time ranges. In particular, we observe short-term effects, which caused an increased message frequency in the first lockdown months and a shifted communication activity during the day in March and April 2020. Moreover, we also see long-term effects of the ongoing pandemic situation until February 2021, which indicate a change of communication behavior towards more regular messaging, as well as a persisting change in activity during the day. The results of our work show that even anonymized chat histories can tell us a lot about people’s behavior and especially behavioral changes during the COVID-19 pandemic and thus are of great relevance for behavioral researchers. Furthermore, looking at the pandemic from an Internet provider perspective, these insights can be used during the next pandemic, or if the current COVID-19 situation worsens, to adapt communication networks to the changed usage behavior early on and thus avoid network congestion.
This paper examines the relationship between time and motion perception in virtual environments. Previous work has shown that the perception of motion can affect the perception of time. We developed a virtual environment that simulates motion in a tunnel and measured its effects on the estimation of the duration of time, the speed at which perceived time passes, and the illusion of self-motion, also known as vection. When large areas of the visual field move in the same direction, vection can occur; observers often perceive this as self-motion rather than motion of the environment. To generate different levels of vection and investigate its effects on time perception, we developed an abstract procedural tunnel generator. The generator can simulate different speeds and densities of tunnel sections (visibly distinguishable sections that form the virtual tunnel), as well as the degree of embodiment of the user avatar (with or without virtual hands). We exposed participants to various tunnel simulations with different durations, speeds, and densities in a remote desktop and a virtual reality (VR) laboratory study. Time passed subjectively faster under high-speed and high-density conditions in both studies. The experience of self-motion was also stronger under high-speed and high-density conditions. Both studies revealed a significant correlation between the perceived passage of time and perceived self-motion. Subjects in the virtual reality study reported a stronger self-motion experience, a faster perceived passage of time, and shorter time estimates than subjects in the desktop study. Our results suggest that a virtual tunnel simulation can manipulate time perception in virtual reality. We will explore these results for the development of virtual reality applications for therapeutic approaches in our future work. This could be particularly useful in treating disorders like depression, autism, and schizophrenia, which are known to be associated with distortions in time perception. For example, the tunnel could be therapeutically applied by resetting patients’ time perceptions by exposing them to the tunnel under different conditions, such as increasing or decreasing perceived time.
After the recent emergence of SARS-CoV-2 infection, unanswered questions remain related to its evolutionary history, path of transmission or divergence and role of recombination. There is emerging evidence on amino acid substitutions occurring in key residues of the receptor-binding domain of the spike glycoprotein in coronavirus isolates from bat and pangolins. In this article, we summarize our current knowledge on the origin of SARS-CoV-2. We also analyze the host ACE2-interacting residues of the receptor-binding domain of spike glycoprotein in SARS-CoV-2 isolates from bats, and compare it to pangolin SARS-CoV-2 isolates collected from Guangdong province (GD Pangolin-CoV) and Guangxi autonomous regions (GX Pangolin-CoV) of South China. Based on our comparative analysis, we support the view that the Guangdong Pangolins are the intermediate hosts that adapted the SARS-CoV-2 and represented a significant evolutionary link in the path of transmission of SARS-CoV-2 virus. We also discuss the role of intermediate hosts in the origin of Omicron.
Since ancient times aging has also been regarded as a disease, and humankind has always strived to extend the natural lifespan. Analyzing the genes involved in aging and disease allows for finding important indicators and biological markers for pathologies and possible therapeutic targets. An example of the use of omics technologies is the research regarding aging and the rare and fatal premature aging syndrome progeria (Hutchinson-Gilford progeria syndrome, HGPS). In our study, we focused on the in silico analysis of differentially expressed genes (DEGs) in progeria and aging, using a publicly available RNA-Seq dataset (GEO dataset GSE113957) and a variety of bioinformatics tools. Despite the GSE113957 RNA-Seq dataset being well-known and frequently analyzed, the RNA-Seq data shared by Fleischer et al. is far from exhausted and reusing and repurposing the data still reveals new insights. By analyzing the literature citing the use of the dataset and subsequently conducting a comparative analysis comparing the RNA-Seq data analyses of different subsets of the dataset (healthy children, nonagenarians and progeria patients), we identified several genes involved in both natural aging and progeria (KRT8, KRT18, ACKR4, CCL2, UCP2, ADAMTS15, ACTN4P1, WNT16, IGFBP2). Further analyzing these genes and the pathways involved indicated their possible roles in aging, suggesting the need for further in vitro and in vivo research. In this paper, we (1) compare “normal aging” (nonagenarians vs. healthy children) and progeria (HGPS patients vs. healthy children), (2) enlist genes possibly involved in both the natural aging process and progeria, including the first mention of IGFBP2 in progeria, (3) predict miRNAs and interactomes for WNT16 (hsa-mir-181a-5p), UCP2 (hsa-mir-26a-5p and hsa-mir-124-3p), and IGFBP2 (hsa-mir-124-3p, hsa-mir-126-3p, and hsa-mir-27b-3p), (4) demonstrate the compatibility of well-established R packages for RNA-Seq analysis for researchers interested but not yet familiar with this kind of analysis, and (5) present comparative proteomics analyses to show an association between our RNA-Seq data analyses and corresponding changes in protein expression.
Despite the fact that mixed-cultural backgrounds become of increasing importance in our daily life, the representation of multiple cultural backgrounds in one entity is still rare in socially interactive agents (SIAs). This paper’s contribution is twofold. First, it provides a survey of research on mixed-cultured SIAs. Second, it presents a study investigating how mixed-cultural speech (in this case, non-native accent) influences how a virtual robot is perceived in terms of personality, warmth, competence and credibility. Participants with English or German respectively as their first language watched a video of a virtual robot speaking in either standard English or German-accented English. It was expected that the German-accented speech would be rated more positively by native German participants as well as elicit the German stereotypes credibility and conscientiousness for both German and English participants. Contrary to the expectations, German participants rated the virtual robot lower in terms of competence and credibility when it spoke with a German accent, whereas English participants perceived the virtual robot with a German accent as more credible compared to the version without an accent. Both the native English and native German listeners classified the virtual robot with a German accent as significantly more neurotic than the virtual robot speaking standard English. This work shows that by solely implementing a non-native accent in a virtual robot, stereotypes are partly transferred. It also shows that the implementation of a non-native accent leads to differences in the perception of the virtual robot.
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.
Though several previous studies reported the in vitro and in vivo antioxidant effect of kinetin (Kn), details on its action in cisplatin-induced toxicity are still scarce. In this study we evaluated, for the first time, the effects of kinetin in cisplatin (cp)- induced liver and lymphocyte toxicity in rats. Wistar male albino rats were divided into nine groups: (i) the control (C), (ii) groups 2,3 and 4, which received 0.25, 0.5 and 1 mg/kg kinetin for 10 days; (iii) the cisplatin (cp) group, which received a single intraperitoneal injection of CP (7.0 mg/kg); and (iv) groups 6, 7, 8 and 9, which received, for 10 days, 0.25, 0.5 and 1 mg/kg kinetin or 200 mg/kg vitamin C, respectively, and Cp on the fourth day. CP-injected rats showed a significant impairment in biochemical, oxidative stress and inflammatory parameters in hepatic tissue and lymphocytes. PCR showed a profound increase in caspase-3, and a significant decline in AKT gene expression. Intriguingly, Kn treatment restored the biochemical, redox status and inflammatory parameters. Hepatic AKT and caspase-3 expression as well as CD95 levels in lymphocytes were also restored. In conclusion, Kn mitigated oxidative imbalance, inflammation and apoptosis in CP-induced liver and lymphocyte toxicity; therefore, it can be considered as a promising therapy.
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.