Digital Library of Gesellschaft für Informatik e.V.
Not a member yet
40758 research outputs found
Sort by
Einsatz von Künstlicher Intelligenz im B2B Customer-Relationship-Management: Identifikation von Herausforderungen und Einsatzpotenzialen in produzierenden kleinen und mittleren Unternehmen
No full textIn einer zunehmend herausfordernden Geschäftsumgebung besteht für kleine und mittlere Business-to-Business Unternehmen die Herausforderung, ihre Geschäftsprozesse und die unternehmensweiten Anwendungssysteme entsprechend der Technologielandschaft anzupassen, um konkurrenzfähig zu bleiben. Der zielgerichtete Einsatz von Künstlicher Intelligenz birgt das Potenzial für Effizienzsteigerung und Unterstützung in verschiedenen Geschäftsbereichen, darunter auch Marketing und Vertrieb. Trotz der zunehmenden Bedeutung von Künstlicher Intelligenz bleibt der Einsatz in kleinen und mittleren Unternehmen oft zögerlich. Das Ziel dieser Studie ist die Herleitung eines ganzheitlichen Ansatzes für eine mit KI-Werkzeugen unterstützte Customer Journey. Diese Arbeit untersucht dafür speziell den potenziellen Einsatz von Künstlicher Intelligenz in der Kundenkommunikation von kleinen und mittleren Unternehmen im produzierenden B2B-Sektor. Basierend auf zehn leitfadenorientierten qualitativen Interviews mit Praktikern aus produzierenden B2B-KMU werden deren spezifische Herausforderungen in der Kundenkommunikation identifiziert und der Einsatz von KI als potenzielle Lösung geprüft. Die Ergebnisse zeigen, dass der Einsatz von KI in der Kundenkommunikation produzierender B2B-KMU Potenzial birgt, um bestehende Herausforderungen zu mindern, wobei die Einsatzmöglichkeiten je nach Datenverfügbarkeit variieren. Die vorliegende Arbeit trägt dazu bei, produzierenden B2B-KMU aufzuzeigen, wie sie Künstliche Intelligenz einsetzen können, um ihre Herausforderungen zu bewältigen und die Kundenkommunikation zu optimieren. In an increasingly challenging business environment, small and medium-sized business-to-business companies face the challenge of adapting their business processes and company-wide enterprise systems in line with the technology landscape in order to remain competitive. The targeted use of artificial intelligence has the potential to increase efficiency and provide support in various business areas, including marketing and sales. Despite the increasing importance of artificial intelligence, its use in small and medium-sized enterprises often remains hesitant. The aim of this study is to derive a holistic approach for a customer journey supported by AI tools. This study specifically examines the potential use of artificial intelligence in the customer communication of small and medium-sized enterprises in the B2B manufacturing sector. Based on ten guideline-oriented qualitative interviews with practitioners from manufacturing B2B SMEs, their specific challenges in customer communication are identified and the use of AI is examined as a potential solution. The results show that the use of AI in the customer communication of manufacturing B2B SMEs holds potential to mitigate existing challenges, although the possible applications vary depending on the availability of data. This study helps to show manufacturing B2B SMEs how they can use artificial intelligence to improve customer communication
Deep Learning zur Wurzelsegmentierung in Minirhizotronbildern mittels Transfer-Learning mit Centerlines-Detektion
No full textDer Aufwand für das Annotieren von Wurzelbildern, die direkt in der Biologie angewendet werden oder als Grundlage für das überwachte maschinelle Lernen eines Modells zur automatisierten Wurzelsegmentierung dienen, ist arbeitsintensiv. In dieser Arbeit wird sich, anhand von binären Centerlines-Annotationen und Wurzelannotationen der Minirhizotronbildern, mit der Frage der Aufwandsreduktion bei der Datenannotation auseinandergesetzt. Mithilfe einer U-Net Architektur wurde zuerst ein Modell zur Centerlines-Detektion trainiert, welches bei der Evaluierung einen F1-Score von 90% erreichte. Durch Transfer-Learning mit dem gelernten Centerlines-Detektion Modell, deren Trainingsdaten einen geringeren Annotationsaufwand besitzen, konnte gezeigt werden, dass die Menge der vollständig annotierten Wurzeln um 87% reduziert werden kann, ohne dass es zu Verlusten in der Genauigkeit (F1-Score von 75%) der Ergebnisse bei Wurzelsegmentierung kommt
A quantum algorithm for solving open system dynamics on quantum computers using noise
No full textWe discuss the importance of non-unitary gates in quantum software engineering by reporting on the results of ‘A quantum algorithm for solving open system dynamics on quantum computers using noise’ published in Physical Review A in December 2023. Non-unitary operations can enable quantum algorithms not possible with purely unitary quantum circuits and are receiving growing interest in the community. One example is the recent proof of quantum advantage for finding local minima under thermal perturbations. Our published work proposes an important building block for non-unitary approaches: An algorithm to approximate complicated non-unitary baths with dedicated bath qubits on a digital quantum computer. We use the results to especially discuss the requirements for quantum software toolkits to support non-unitary algorithms
Privacy Visualizations in F-Droid: Introducing an Interactive Privacy Visualization Enabling Users To Reflect on Their Privacy Needs
No full textF-Droid is an app store that focuses on providing a platform for free open-source software for Android and is considered the main alternative to Google’s Play Store. This work aims to tackle two challenges of the F-Droid app store in terms of visualizing privacy-related information of apps. First, F-Droid’s privacy metadata can be limited, unclear, and technically inaccessible for most users. Second, these apps’ privacy parameters miss design features that enable users to reflect on their privacy concerns in an informative and interactive way. To address these challenges, this work used human-centered design methodology to develop and evaluate four low-fidelity prototypes based on existing research in the field of privacy visualization. These prototypes were discussed with domain experts revealing that users’ preferences for the design of privacy visualizations vary in regard to their privacy concerns. Experts who were specifically concerned with companies like Google tended to prefer more informative designs, while privacy advocates who were concerned with government surveillance preferred less cognitively demanding designs. Using these insights, the most promising prototype was refined and implemented as an interactive high-fidelity prototype in Neo Store, a modern F-Droid client. All in all, this work represents a new direction in visualizing privacy-related information in app stores, aimed at raising users’ awareness of their privacy needs and concerns
Adaptive sorting for large keys, strings, and database rows
No full textSorting a database table may require expensive comparisons, e.g., due to column count or column types such as long or international strings. Therefore, optimizing the count and cost of comparisons is important. Adaptive sort algorithms avoid comparisons by exploiting pre-existing order in the input. If keys happen to be sorted or reverse-sorted, comparisons suffice, in contrast to comparisons in the expected and worst cases. Ideally, adaptive sorting ensures graceful degradation from the best case to the expected case, e.g., by merging sorted runs pre-existing in the input. Adaptive sorting has proven successful for integer keys but not for large keys, e.g., when comparing symbols or characters in text strings, bytes or words in binary strings, or column values in database rows. On the other hand, using longest common prefixes or offset-value codes, sorting strings with characters, bytes, or columns can be limited to comparisons. If those comparisons are the dominant cost of sorting, e.g., due to interpreted execution of predicates etc., the cost of sorting is linear in the input size and, in fact, equal to the cost of verifying a claimed and correct sort order. By leveraging and combining a variety of techniques, some old and proven in practice, some new yet equally sound, we introduce sorting techniques that are efficient, scalable, and adaptive; that degrade gracefully from to comparisons of strings or of database rows; and that guarantee at most comparisons of bytes or of column values in the worst case. We offer algorithm analyses and experimental results to test our hypotheses and support our claims
Trash Talk
No full textGelber Sack sei Dank: In Sortieranlagen für Kunststoff-Recycling landen täglich tonnenweise Abfälle – und bei Weitem nicht nur Kunststoffe. Je präziser Materialien erkannt und sortiert werden, desto umweltfreundlicher können sie recycelt werden. Hier spielt künstliche Intelligenz eine Schlüsselrolle. Johannes Laier von WeSort.AI hat einen Algorithmus entwickelt, der eine Vielzahl von Objekten in Stoffströmen erkennt, Marken von PET-Flaschen ausliest und sogar CO₂-Fußabdrücke berechnen kann. Im Interview erklärt er, wie genau das funktioniert und welche Entwicklungen uns in Zukunft erwarten
Transactional YCSB: Benchmarking ACID-Compliant NoSQL Systems with Multi-Operation Transactions
No full textNoSQL systems are popular because of their flexible data models and focus on availability, scalability, and fault tolerance. They often have loosened ACID guarantees to achieve these goals. To also support use cases that rely on ACID transactions, some existing NoSQL systems introduced ACID compliance after their release, and new NoSQL systems are created to support ACID compliance from the ground up. A benchmark that supports transactions is required to compare the performance of these ACID-compliant NoSQL systems. The Yahoo! Cloud Serving Benchmark (YCSB) is the most used benchmark for NoSQL systems but does not support transactions. YCSB+T, an extension of YCSB, introduces support for transactions into the YCSB, but only for transactions consisting of a single operation. A further extension of YCSB is required to support the performance evaluation of workloads containing transactions that consist of multiple operations. This paper introduces Transactional YCSB, an extension of the benchmarking framework YCSB that enables the evaluation of ACID-compliant NoSQL systems for workloads consisting of multi-operation transactions. Further, the paper evaluates the ACID-compliant NoSQL systems FoundationDB, MongoDB, and OrientDB using the developed YCSB extension
Modeling and Simulation of MPI Communication Dynamics in InfiniBand Networks Using SimGrid
No full textInfiniBand is a network architecture introduced to fulfill the need for high-bandwidth and low latency connectivity in both I/O and inter-processor communication. The widespread support from most vendors has placed InfiniBand as a key solution to enhance connectivity speed in next-generation HPC platforms. Meanwhile, Modeling and simulating high-performance interconnect (InfiniBand specifically) is crucial for the design of future HPC platforms. Specifically modeling the dynamic behavior of communication libraries such as MPI (Message Passing Interface) in how they manage internal communication modes based on message sizes on top of InfiniBand networks helps to achieve more accurate predictions. In this paper, we look into the step-by-step process of modeling and simulating Point-to-Point MPI communication on the top of InfiniBand interconnect network. This process involves exploring how closely we can align the communication behavior of the InfiniBand network with piece-wise linear models. For that, we employ these piece-wise linear models in SimGrid via SMPI as hybrid LogGPS models. The similarity between communication costs in the simulation and real-world measurements indicates that the selected communication model can be used in modeling of larger networks (e.g., using InfiniBand switches)
Management und FAIRe Bereitstellung von multi-dimensionalen analysefertigen Rasterdaten mittels dynamischer Webdienste
No full textAls Ressortforschungseinrichtung des Bundesministeriums für Ernährung und Landwirtschaft (BMEL) unterhält das Julius Kühn-Institut (JKI) eine Geodateninfrastruktur (JKI-GDI), die um Big-Geodata-Komponenten erweitert worden ist. Der Beitrag skizziert das Transformationsergebnis von einer zentralisierten lokalen behördlichen GDI zu einer dezentralisierten GDI mit Big-Geodata-Komponenten. Der JKI-Datacube repräsentiert dabei einen entscheidenden Baustein der JKI-GDI. Die Bedeutung ergibt sich aus der performanten Speicherung multidimensionaler Rasterdaten sowie der Datenbereitstellung durch standardisierte Webdienste. Mit der JKI-GDI steht eine Arbeitsumgebung zur Verfügung, mit der Big-Geodata-Daten-integrationsoperationen und Analysen operativ durchgeführt werden können. Die Webdienst-basierten Funktionalitäten ermöglichen eine dynamische Datenbereitstellung entsprechend den FAIR-Prinzipien, was anhand einer parzellenspezifischen Datenintegration veranschaulicht wird
Exploration of Efficient Computation for Trajectory Planning via Fixed-Point Arithmetic
No full textAs the computational complexity of autonomous driving modules increases, reducing computational energy consumption becomes increasingly important. Additionally, to enable vehicles to make rapid decisions in dynamic environments, real-time processing of these functions is essential. Fixed-point arithmetic, compared to floating-point arithmetic, involves simpler arithmetic operations, requiring fewer logic gates and transistors, thus leading to lower hardware resource consumption and energy usage. Reducing the bit width of fixed-point numbers to enhance SIMD computation efficiency could be highly advantageous for autonomous driving modules. However, using smaller bit-width data to represent numbers typically results in some loss of precision. Maintaining precision within a tolerable range is crucial. This paper explores the applicability of fixed-point arithmetic for trajectory planning. The experiments analyze how the fractional bit width of fixed-point numbers affects precision loss in trajectory calculations. The results show that using a 32-bit fixed-point representation ensures that the calculated position of the trajectory points stay within 2 cm accuracy over a 50 m trajectory. These experimental results can guide the design of FPGA ALUs with specific bit widths, thereby improving computational efficiency