Variational inference for Markovian queueing networks

We consider networks of queues in discrete time, where the steady state distribution can be computed explicitly in closed form (product form networks): (i) Closed cycles and open tandems of single server FCFS Bernoulli nodes with state dependent service probabilities, where customers flow linearly, (ii) networks of doubly stochastic and geometrical queues (which are discrete time analogues of Kelly’s symmetric, resp. general, servers), where customers of different types move through the network governed by a general routing mechanism and request for service according to general, resp. geometrical, distributions, (iii) networks with batch movements of customers and batch service, where the service and routing mechanism is defined via an abstract transition scheme.

DASH is a new parallel programming model for HPC which is implemented as a C++ template library on top of a runtime library implementing various PGAS (Partitioned Global Address Space) substrates. DASH’s goal is to be an easy to use and efficient way to parallel programming with C++. Supporting software tools is an important part of the DASH project, especially debugging and performance monitoring. Debugging is particularly necessary when adopting a new parallelization model, while performance assessment is crucial in High Performance Computing applications by nature. Tools are fundamental for a programming ecosystem and we are convinced that providing tools early brings multiple advantages, benefiting application developers using DASH as well as developers of the DASH library itself. This work, first briefly introduces DASH and the underlying runtime system, existing debugger and performance analysis tools. We then demonstrate the specific debugging and performance monitoring extensions for DASH in exemplary use cases and discuss an early assessment of the results.

Chemiresistive gas sensors are an important tool for monitoring air quality in cities and large areas due to their low cost and low power and, hence, the ability to densely distribute them. Unfortunately, such sensor systems are prone to defects and faults over time such as sensitivity loss of the sensing material, less effective heating of the surface due to battery loss, or random output errors in the sensor electronics, which can lead to signal jumps or sensor stopping. Although these defects usually can be compensated, either algorithmically or physically, this requires an accurate screening of the entire sensor system for such defects. In order to properly develop, test, and benchmark corresponding screening algorithms, however, methods for simulating gas sensor networks and their defects are essential. In this work, we propose such a simulation method based on a stochastic sensor model for chemiresistive sensor systems. The proposed method rests on the idea of simulating the defect-causing processes directly on the sensor surface as a stochastic process and is capable of simulating various defects which can occur in low-cost sensor technologies. The work aims to show the scope and principles of the proposed simulator as well as to demonstrate its applicability using exemplary use cases.

Physical attacks, such as fault attacks, pose a decisive threat for the security of devices in the Internet of Things. An important class of countermeasures for fault attacks is fault tolerant software that is applicable for systems based on COTS hardware. In order to evaluate software countermeasures against fault attacks, fault injection is needed. However, established fault injection approaches require manufactured products or hardware details (e.g. netlists, RTL models), which are not available when using COTS hardware. In this paper, we present a QEMU-based fault injection platform that supports commercial COTS processors that are widely-used in the embedded domain. This framework allows a system-level analysis of software countermeasures by featuring the simulation of high-level hardware faults targeting, for example, memory cells, register cells, or the correct execution of instructions. The framework supports the generation of realistic fault attack scenarios. We illustrate the practicability of the approach by presenting two exemplary use cases.

In IoT environments, data are collected by many distinct devices, at the periphery, so that their feature-sets can be naturally endowed with a faceted structure. In this work, we argue that the IoT requires specialized ML models, able to exploit this faceted structure in the learning strategy. We demonstrate the application of this principle, by a multiple kernel learning approach, based on the exploration of the partition lattice driven by the natural partitioning of the feature set. Furthermore, we consider that the whole data management, acquisition, pre-processing and analytics pipeline results from the composition of processes pursuing different and non-perfectly aligned goals (most often, enacted by distinct agents with different constraints, requirements competencies and with non-aligned interests). We propose the adoption of an adversarial modeling paradigm across the overall pipeline. We argue that knowledge of the composite nature of the learning process, as well as of the adversarial character of the relationship among phases, can help in developing heuristics for improving the learning algorithms efficiency and accuracy. We develop our argument with reference to few exemplary use cases.

Energy systems worldwide face transitions by the integration of renewable generation sources which are a major driver to reduce carbon emissions and fulfill decarbonization targets. To quantify mid- and long-term effects of these transitions, energy system optimization models are applied. These models can support the design of efficient policies and the development of suitable technologies. However, modeling each country individually requires computational power, sufficient data, and human processing time for the validation and result evaluation. This thesis provides an efficient framework for the comparison of countries’ decarbonization pathways. Instead of increasing the computational power with brute force, the key principal is a data-driven approach accelerating the preparation and evaluation of energy system models. For this purpose, clustering algorithms are applied in two hierarcical stages: the first stage summarizes countries in global energy system archetypes to assess transitions for a reduced number of prototypical countries. The second stage clusters spatially highly resolved data to generate suitable regions for a multi-region model considering the differing spatial potentials of renewable generation. Both approaches are validated by three exemplary use cases: 1) grid topology, 2) green hydrogen production, and 3) coal phase-out. The developed approach is based on a global data basis combining socio-economic, geographic, and energy data including highly resolved geospatial data. This data basis is required for both clustering algorithms. The archetype clustering uses an extended k-Means algorithm. In the second stage, Ward’s method is implemented to cluster the spatial data. To avoid the curse of dimensionality by the high spatio-temporal resolution of hourly renewable generation profiles, Dynamic Time Warping and Principal Component Analysis reduce the time dimension in the clustering. For the first and third use case, the framework processes OpenStreetMap data to synthesize the existing grid structure and identify coal mining areas. The hydrogen use case uses a break-even price approach. The countries are classified in 15 archetypes, which is in a similar range as the definition of subregions by the United Nations. Compared to these regions, the classification in archetypes considers countries on different continents and thereby represents the energy system characteristics on average 30% better. Modeling an 80% decarbonization scenario between 2015 and 2045 for all archetypes shows the fundamental challenge that huge investments in currently less developed countries are needed. Furthermore, the modeling results confirm that the decarbonization pathway of countries is 10-30% closer to countries within the same archetype than to countries within the same geographic region. The use cases apply the developed framework from archetype clustering to regional clustering. For all three use cases, the archetypes lead to a selection of countries with suitable characteristis. Regarding the grid, Denmark, representing a country of a high renewable share, foresees fewer changes in its grid topology than Morocco. Second, Saudi Arabia is highly attractive for green hydrogen production with a low break-even price of 1€/kg in the context of a decarbonized system. Last, coal-dominated regions in South Africa face significant challenges since a coal phase-out shifts the generation to other regions with good renewable conditions. This thesis contributes to the worldwide application of energy system models which are important to determine cost-optimal decarbonized energy systems. The implementation of data-driven clustering methods simplifies and accelerates the modeling of each country globally. Thereby, modelers can use the data analysis as a first indication and focus on the model or its evaluation, policymakers can compare countries, or even regions within countries, and technology companies are able to quickly assess markets for their products.

SIoT and FC are two stand-alone technological paradigms under the realm of the Future Internet. SIoT relies on the self-establishment and self-management of inter-thing social relationships, in order to guarantee scalability to large IoT networks composed of both human and non-human agents. FC extends cloud capabilities to the access network, in order to allow resource-poor IoT devices to support delay-sensitive applications. Motivated by these complementary features of the SIoT and FC models, in this article we propose their integration into the novel paradigm of the SoFT. Specifically, we provide the following three main contributions. After describing the SoFT paradigm, we discuss its introduction through a number of exemplary use cases. We describe the architecture and the main resource-management functions of the resulting virtualized SoFT technological platform. It merges the physical things at the IoT layer and their virtual clones at the Fog layer into a cyber-physical overlay network of social clones. As a proof-of-concept, we present the simulated performance of a small-scale SoFT prototype, and compare its energy-vs.-delay performance with the corresponding one of a state-of-the-art virtualization-free technological platform, which relies only on device-to-device (D2D) inter-thing communication.

The current transformation process of how energy is supplied attracts great interest from many different market players. As a consequence, many proprietary solutions for “smart” energy applications are flooding the market. This turns out to be rather a problem than part of the solution for the systematic development of future energy grids. Additionally, the absence of necessary standards blocks further developments that enable the creation of novel, market-driven and hybrid control solutions. To overcome these problems, we suggest a standardized control approach for hybrid energy systems by means of a so called Energy Option Model (EOM). This unifying model and the therewith developed decision support system provides the necessary technical understanding and the economic assessment options for network-connected energy conversion systems. Thus, it can be used for single on-site systems as well as for aggregated systems that are controlled in centralized or decentralized manner. This paper presents and discusses exemplary use cases for our EOM that illustrate the centralized as well as the decentralized use of our approach within hybrid energy systems. Overall, we believe that the EOM represents the key approach for a further systematic development of an open hybrid energy grid.

The International Telecommunication Union (ITU) has proposed immersive communication scenarios, such as Extended Reality (XR), as exemplary use cases for 6G networks, intensifying the concurrent demands for low latency and high throughput. Consequently, the design of 6G solutions should not only focus on how to further increase data transmission rates but also on how to transmit the most valuable data to utilize limited communication resources more efficiently, thereby achieving superior system performance. In this paper, we aim to enhance the XR user experience in terms of timeliness and resolution under resource constrained scenarios. On the one hand, we increase the wireless communication system transmission capability via multi-connectivity technology. On the other hand, we reduce the source transmitting data via transmitting valuable source data. Specifically, we design a value-driven multi-connectivity transmission scheme that jointly optimizes source decision and resource allocation for XR services, aiming to efficiently utilize the resources for serving XR services. Furthermore, we design the Value of Video Data (VoVD) as the metric to comprehensively measure the XR performance, which balances timeliness and resolution. We also propose a Cascaded Dimension-Refined Reinforcement Learning (CDRRL) algorithm, which delivers a stable, low-complexity solution. The numerical results confirm the significant performance improvements and algorithm stability.

AI-powered knowledge graphs: Accelerating simultaneous engineering in the automotive digital era

With the advent of 'heavy Artificial Intelligence' – big data, deep learning, and ubiquitous use of the internet, ethical considerations are widely dealt with in public discussions and governmental bodies. Within Computational Paralinguistics with its manifold topics and possible applications (modelling of long-term, medium-term, and short-term traits and states such as personality, emotion, or speech pathology), we have not yet seen that many contributions. In this article, we try to set the scene by (1) giving a short overview of ethics and privacy, (2) describing the field of Computational Paralinguistics, its history and exemplary use cases, as well as (de-)anonymisation and peculiarities of speech and text data, and (3) proposing rules for good practice in the field, such as choosing the right performance measure, and accounting for representativity and interpretability.

Cold spraying has emerged as a promising technique for the repair of metallic components. For controlled material deposition, manipulation of the cold spray gun is performed by industrial robots. Such robot-guided cold spraying provides flexible control and automation of the entire process. To enable effective and material-efficient material deposition at specified repair locations, this work proposes a method for automated planning of cold spray paths and trajectories. The method begins with the extraction of the volume to be filled by comparing the nominal and actual component. To produce the extracted volume, it is divided into suitable adaptively curved layers and converted into point clouds for path planning. The cold spray path is then converted into a trajectory by adding a suitable velocity distribution for the spray velocity to produce the required locally varying layer thickness. To validate the suitability of the path and trajectory, a simulation of the material deposition is performed. In addition, the implementation of the entire method is demonstrated by exemplary use cases. The results demonstrate that the proposed method enables successful automated path and trajectory planning, both contributing to the overall goal of automated repair of damaged components by cold spray.

Social media services like Twitter churn out user-generated content in vast amounts. The massive availability of this kind of data demands new forms of analysis and visualization, to make it accessible and interpretable. In this article, we introduce Twista , an application that can be used to create tailored tweet collections according to specific filter criteria, such as the occurrence of certain keywords or hashtags. Once the tweet collection has been created, Twista calculates basic statistics, e.g. the average tweet length or the most active user. Furthermore, the application can perform basic sentiment analysis, analyze tweets with regard to their date of publication, and analyze the communication between different Twitter users. The results of these analyses are visualized by means of the data driven documents toolkit ( d3.js ) and can be viewed directly in the browser, or are available for download in PDF and JSON format. We also present three exemplary use cases that illustrate the possible use of Twista for different scenarios.

A large number of stochastic networks including loss networks and certain queueing networks have product-form steady-state probabilities. However, for most practical networks, evaluating the system performance is a difficult task due to the presence of a normalization constant. We propose a new framework based on probabilistic graphical models to tackle this task. Specifically, we use factor graphs to model the stationary distribution of a network. For networks with arbitrary topology, we can apply efficient message-passing algorithms like the sum-product algorithm to compute the exact or approximate marginal distributions of all state variables and related performance measures such as blocking probabilities. Through extensive numerical experiments, we show that the sum-product algorithm returns very accurate blocking probabilities and greatly outperforms the reduced load approximation for loss networks with a variety of topologies. The factor graph model also provides a promising approach for analyzing product-form queueing networks.

Electronic Health Records: Promises and Realities: A 3-Part Series Part I: The Digital Sea Change, Ready or Not