1194 research outputs found
Sort by
Open GIS-to-BIM-Converter - LKMap 2025 Data to IFC 4.3 Data
Get PDFUnderground utility networks in Switzerland are typically managed using normed formats such as the SIA 405-INTERLIS standard within 2D Geo-Information Systems (GIS). However, due to the increasing demand for 3D representation, particularly in the context of Building Information Modeling (BIM), there is a need to represent these networks in a 3D format. This research project focuses on the SIA405_LKMap model, which describes the minimum information required for cadastral registers and is currently being revised to support 3D Geo-Information. The research presents an evaluation of the recommended changes in the revision of SIA405_LKMap and a Python Command Line Interface (CLI) prototype that enables the conversion of SIA405_LKMap data into valid 3D data in the Industry Foundation Classes (IFC) format.
The resulting prototype has a CLI that requires users to input a INTERLIS LKMap transfer data file and a reference null point, while also providing other optional arguments. The provided arguments are validated, and the conversion process is performed using Python libraries such as Pyogrio and IfcOpenShell. The resulting 3D data includes element attributes as IfcPropertySets and the visualization of geometric uncertainties.
The prototype provides a strong basis for future development. The outlook includes additional improvements, such as optimizing geometric representations and adding configurability and extensibility features for users. Overall, this thesis demonstrates the feasibility of converting INTERLIS LKMap data to the IFC 4.3 format, opening up possibilities for improved representation and visualization of underground utility networks in the field of BIM.
Keywords: 2D to 3D conversion, INTERLIS, LKMap, IFC, Python CLI, GeoPackage, Building Information Modeling (BIM)
TypeSearch: Type-Directed API Search For All
Get PDFSoftware developers spend a lot of their time finding and composing pre-existing functions from various libraries. Almost all developers today use general-purpose search engines for this search. Specialized search engines such as Hoogle additionally use type information to improve this search, and have been successful for some typed functional programming languages. The options currently available for type-directed search for mainstream object-oriented languages is limited. Existing approaches for these languages do not have first-class support for subtyping or parametric polymorphism.
The splitting and composition of a desired functionality into and from a number of pre-existing functions is also a task that needs to be done manually. In this Master’s Thesis we present a proof-search-based approach to type-directed search with first-class support for subtyping, parametric polymorphism, splitting, and composition. The approach is language agnostic, and can be specialized to simultaneously support multiple typed object-oriented languages. Given that most mainstream languages fall under this category, this approach would extend the benefits of type-directed search to the majority of programmers. As a proof of concept, we provide a running implementation of the core language-agnostic approach and extend it to support the Java programming language. Further extensions would allow the tool to simultaneously support multiple programming languages using the same query syntax
Interaktion in natürlicher Sprache
Get PDFIm Laufe dieser Masterarbeit wird ein codierter Prototyp für einen persönlichen intelligenten
Sprachassistenten entwickelt, der insbesondere für ältere, in einem Pflegeheim lebende
Menschen gedacht ist. Anhand der Entwicklung wird ein möglichst umfassendes Verständnis
der Interaktion zwischen Menschen und Computern mittels natürlicher Sprache angestrebt.
Dazu werden zwischenmenschliche Dialoge erhoben und ausgewertet, Dialogstrukturen
analysiert und die Regeln des Turn-Taking untersucht. Diese Aspekte sind zentrale
Bestandteile in der Entwicklung und Nutzung von sprachbasierten Assistenten.
Da zu Beginn der Arbeit kein bestimmter Anwendungsfall vorhanden war, mussten ein
sinnvoller Einsatz und eine Gruppe von Nutzenden gesucht und gefunden werden. Das
Vorgehensmodell LeanUX wurde angewendet, um in kurzer Zeit Anwendungsfälle auf ihren
Nutzen überprüfen zu können. Ausgehend von technologischen Möglichkeiten soll flexibel auf
Bedürfnisse eingegangen und die Grenzen der aktuellen Technologien sollen aufgezeigt
werden.
Diese Arbeit befasst sich ebenfalls mit dem Prozess der Produktentwicklung, in dem künstliche
Intelligenz Aufgaben des Menschen übernimmt. Es wird beschrieben, weshalb bekannte
Vorgehensmodelle und Artefakte wie Persona, Journey und Dialogmodelle wenig zielführend
sind. Um die aufgeworfenen Forschungsfragen beantworten zu können, werden Experimente
durchgeführt, in welchen Menschen die Aufgabe eines Systems übernehmen. Über fünf Zyklen
hinweg wird der Mensch sukzessive durch ein System ersetzt, um am Ende einen codierten
Prototyp unter realen Bedingungen mit Nutzenden auf seine Nutzbarkeit und Nützlichkeit zu
überprüfen. Dessen Entwicklung dient als praktisches Beispiel und soll es ermöglichen, die
theoretischen Erkenntnisse in die Praxis umzusetzen und zu testen. Die Entwicklung und die
Analyse des Prototyps bieten wertvolle Einblicke in die Herausforderungen und Potenziale der
Sprachinteraktionstechnologie. Dabei werden auch ethische Aspekte in der Interaktion von
Menschen mit künstlicher Intelligenz untersucht.
Zusammenfassend soll mit dieser Arbeit ein elementarer Beitrag zum Verständnis der
menschzentrierten Entwicklung von Sprachinteraktionssystemen geleistet werden. Es werden
detaillierte Analysen der Dialogstrukturen und Interaktionsmechanismen sowie praktische
Ansätze für die Gestaltung nutzungsfreundlicher und ethisch verantwortungsvoller
Sprachassistenten aufgezeigt. Die Ergebnisse dieser Arbeit sind daher von grosser Relevanz
für die Forschung und Entwicklung im Bereich der Mensch-Computer-Interaktion im Umgang
mit künstlicher Intelligenz
Natural Language to GraphQL
Get PDFIn-context learning enables large language models (LLMs) to comprehend and respond based on the context of the input, allowing them to adapt to a wide range of domain-specific tasks without additional training. However, they have notable limitations in accessing real-time or domain-specific data because they operate primarily on the knowledge they were trained on. Natural language-to-SQL (NL-to-SQL) systems offer a practical solution, enabling LLMs to transform natural language into SQL commands. This makes data accessible to people without technical expertise. GraphQL, having emerged as a flexible alternative to REST, enables software clients to specify the precise data they require from an API based on a schema.
Based on this research, the project aimed to evaluate the best practices for building a natural language interface for GraphQL by adapting the concepts of NL-to-SQL to the GraphQL domain. By understanding the concepts and evaluating a few optimal strategies, a strong foundation was laid for future development.
Upon starting the project, in-depth research was conducted to understand the current state-of-the-art, focusing on NL-to-SQL due to its similarities and substantial literature available. Key findings were documented, and possible strategies were defined. After a proof-of-concept, two strategies, with a few sub-variants, were established for implementation. The first strategy solely relies on the capabilities of LLMs to directly follow instructions and, using in-context learning, enhance the prompt with relevant samples. The second strategy focuses on using entity extraction to identify entities in the user's question, match them to the schema, and then build the operation based on an abstract syntax tree. During implementation, a third strategy combining the benefits of both previous approaches evolved that overcomes the context size limitation of LLMs. In order to measure the strategies' performance in various metrics, an evaluator was built to efficiently test different implementations against a test set. The latter was inspired by the Spider dataset, a widely used benchmark for NL-to-SQL solutions.
The evaluation has seen 7 different LLMs tested against the 4 most mature variants. The results were analyzed to determine the best-performing combinations. While the first strategy showed promising results for simple test cases, it demonstrated limitations in terms of quality and consistency for more complex ones. The best performing combination uses entity extraction and algorithmic query generation, which is capable of correcting intermediate errors and always produces valid output, making it reliable enough to be used in experimental environments. In general, OpenAI models (GPT-4) are reliable in following instructions, while open source models (Llama3, Mistral) have trouble consistently generating valid structures such as JSON. Hallucinations, though occurring on both OpenAI and open source models, can be drastically reduced with prompt engineering, making the output more consistent. However, both strategies are limited by the context size of the LLM used, making them cost-inefficient or even non-processable for large schemas. To overcome this limitation, future research should focus on an iterative entity extraction approach, enabling large schemas to be processed. Additionally, the shot sampling process can be optimized to further benefit from the LLM's in-context learning capabilities
Limbic Chair Anwendung
Get PDFEinleitung: Der Limbic Chair ist eine innovative
Sitzlösung von Limbic Life. Er nutzt
Sensortechnologie, um durch Bewegungstracking die
Interaktion mit dem Benutzer zu verbessern. Diese
Bachelorarbeit untersucht die Entwicklung eines
fortschrittlichen Softwaresystems, das die
Funktionalitäten des Limbic Chairs durch die Nutzung
seiner integrierten Sensoren erweitert. Der Stuhl ist
mit sechs Sensoren ausgestattet, die verschiedene
Beinbewegungen überwachen und aufzeichnen.
Insbesondere erfassen die linken und rechten Pitch-
Sensoren die vertikalen Bewegungen der Beine,
während die linken und rechten Yaw-Sensoren die
seitlichen Beinbewegungen verfolgen. Darüber
hinaus messen die Roll-Sensoren auf beiden Seiten
die Rotationsbewegungen des Stuhls. Ziel dieses
Projekts ist es, diese Sensoreingaben in eine
Softwareplattform zu integrieren, die ein erweitertes
Benutzererlebnis bietet und den Weg für innovative
Anwendungen in Bereichen wie Gesundheit und
Gaming ebnet. Der Entwicklungsprozess, die
Systemarchitektur und die potenziellen Anwendungen
des erweiterten Limbic Chairs werden in dieser Arbeit
aufgezeigt.
Vorgehen / Technologien: In der heutigen Zeit sitzen
viele Menschen aufgrund ihrer beruflichen und
privaten Aktivitäten zu lange, was nachweislich
negative Auswirkungen auf die Gesundheit hat, wie
Rückenprobleme und schlechte Körperhaltung. Diese
Bachelorarbeit zielt darauf ab, dieses Problem
anzugehen und eine Lösung zu entwickeln, die zu
mehr Bewegung im Sitzen anregt.
Eine umfassende Recherche und Analyse wurden
durchgeführt, um die gesundheitlichen Auswirkungen
von übermässigem Sitzen zu untersuchen und
bestehende Lösungen zur Bewegungsförderung am
Arbeitsplatz zu analysieren. Basierend darauf wurde
der Movement Score entwickelt, der verschiedene
Parameter zur Berechnung der Bewegung des
Nutzers heranzieht. Ein Algorithmus berechnet den
Score kontinuierlich basierend auf den getrackten
Bewegungen.
Zur Erhöhung der Attraktivität des intelligenten Stuhls
wurden zwei interaktive Bewegungsspiele entwickelt,
die den Stuhl als Controller nutzen. Im Raketen-Spiel
steuert der Benutzer eine Rakete, während er im
Jump-and-Run-Spiel einen Charakter durch eigene
Bewegungen im Stuhl steuert. Diese Spiele sollen zu
mehr Bewegung motivieren und gleichzeitig Spass
bieten.
Ergebnis: Das Usability-Testing zeigte, dass Benutzer
sich deutlich mehr bewegten, sobald sie sich des
Movement Scores bewusst waren. Der Score stieg
schnell an, da Benutzer motiviert waren, ihren
Punktestand zu verbessern. Die Nutzung der
Bewegungsspiele führte zu gemischten Ergebnissen:
Einige spielten bis zu 15 Minuten, andere nur kurz.
Ein weiterer positiver Effekt war das gesteigerte
Bewusstsein der Benutzer für ihre täglichen
Aktivitäten. Viele versuchten, sich auch im Sitzen
mehr zu bewegen. Bei Aufgaben mit hoher
Konzentration wurde der Movement Score jedoch
häufig ignoriert, während bei leichteren Tätigkeiten
wie Meetings eher passive Bewegungen in den Alltag
integriert wurden
Document Management using Large Language Model
Get PDFInitial Situation:
The management and retrieval of documents can be challenging due to the unstructured nature
of their content. Traditional search methods, which rely on document names, are often inefficient
and ineffective. With the rise of Transformers-based Large Language Models (LLMs), which sig-
nificantly enhance our everyday tasks, there is an opportunity to improve the search and man-
agement of these documents. The integration of LLMs can transform unstructured data into
structured metadata, making documents more accessible and organized.
Objective:
Within the scope of this project, we aim to build a Single Page Application of this type. We created this prototype application, which is able to store, read, and process PDF documents with unstructured data and generate metadata using an LLM. This metadata improves document search and
management, as well as streamlines repetitive tasks such as sending an email reminder to a
customer. The prototype is designed to be easily expandable to facilitate the continuous development and implementation of new features.
Conclusion:
In this project, we developed a prototype application utilizing React and TypeScript for the frontend while using NestJS with TypeScript for the backend. We integrated external services such as Hugging Face and Zapier. The application enables users to upload PDF documents, extract
a suitable title, summary, and tags using a Large Language Model (LLM), and search for docu-
ments based on this metadata. Additionally, it includes a feature for sending email reminders
to customers for specific files. Designed with scalability in mind, the application allows for the
easy addition of new features in the future, such as integrating various file types like pictures,which could utilize the LLM. This proof of concept demonstrates the potential of using LLMs to enhance document management and retrieval
Consilium
Get PDFProject management and organizational software is a complicated domain and already a well-saturated and competitive market.
Due to the various levels of technical affinity, user-friendliness is a key aspect for software, trying to gain a foothold in this sector.
Especially for smaller ventures, and freelancers, project management and administrative tasks are often not their main activities, but something keeping them from spending more time doing what they really want.
On top of that, project management, creating offers and invoices are often only seen as a means to an end and not something to be enjoyed.
Therefore, project management software should make the lives of their users easier, by reducing the overhead to a minimum.
Tools in this domain should offer broad functionality while not getting overly complicated to use.
This thesis identifies smaller ventures and freelancers as the perfect market segment to target with software tailor-made for their needs through a market analysis.
Consilium addresses their needs by uniting different tasks, previously divided across multiple tools into one end-to-end software solution.
While multiple freelancers and small ventures have previously been creating their offers with Word, tracking their efforts with Excel and making their notes in a text editor, they can now just use Consilium.
Consilium identified the synergies that can be used between those various administrative tasks to make them simpler.
By introducing innovative concepts, Consilium managed to become a well-rounded administrative tool, that can learn with the users, offering functionality from creating offers to creating invoices, while maintaining sufficient depth to model even complex workflows.
Consilium is integrated into the AWS cloud and offers uncomplicated login functionality allowing potential users to sign up in seconds, using social logins.
While primarily targeting freelancers, Workspaces are implemented, allowing users to collaborate.
On top of all the functionality offered, Consilium maintains a high level of customizability, allowing for custom branding on exported documents and content to be added using a Rich text editor, which is then translated and exported into a PDF.
Consilium manages to reduce the administrative overhead significantly by building smart workflows, through clever reuse of data.
Early user tests show, Consilium is already in a state, which helps freelancers improve their productivity, by streamlining processes, showing significant improvements, compared to their previous processes.
Consilium has the potential to become even more.
By introducing a customer-view Consilium could become a marketplace, where users could tender projects and freelancers could offer their work.
Here, a match-making algorithm could achieve groundbreaking results for freelancers and make finding someone to implement your project extremely easy.
Another interesting lane to go down would be to add the feature where customers could approach freelancers with a potential project description which through conversational artificial intelligence could be translated into an offer draft for the service provider
Training a simulated drone with deep reinforcement learning
Get PDFThis report details the comprehensive planning, execution, and evaluation of a solodeveloped project in the field of artificial intelligence (AI) and software engineering.
The project, focused on implementing a drone capable of autonomous navigation and delivery, follows the Agile Scrum methodology for project management. The report covers various aspects, including risk management, iterative development processes, and the application of the Soft Actor-Critic (SAC) algorithm for AI training.
A key aspect of the report is the detailed description of the SAC algorithm, which outlines the mathematical principles of this algorithm in an understandable way.
The development lifecycle is outlined through a series of sprints, each encompassing planning, review, and retrospective meetings. The report underscores the importance of risk management, categorizing and addressing potential challenges throughout the project.
Detailed insights into the project’s structure, from backlogs to sprints, provide a transparent view of the iterative development process. The challenges faced, such as lack of experience with AI and Unreal Engine, are mitigated through proactive measures, including research and adaptation.
The long-term plan, presented in a Gantt chart, highlights key milestones, phases, and features. Primary and secondary features, along with epics and milestones, are meticulously defined, allowing for a clear understanding of project progress.
The report concludes with a thorough retrospective, highlighting successes, areas for improvement, and personal reflections. Key achievements include overcoming the hyperparameter problem and implementing the SAC algorithm to control the drone
Wi-Fi Security Threats - an Integrative Review
Get PDF"What are possible threats against Wi-Fi infrastructure?" is the main research question this integrative literature-review wants to find an answer to. Whether those threats are adequately dealt with, and how impactful the real-world implications appeared to be, are additional research goals. It was aimed to classify and illustrate those threats and attacks in an attack tree, demonstrating the possible attack vectors and paths.
The literature-review is based on scientific sources (papers, articles) in the field of vulnerabilities in Wi-Fi security, protocol weaknesses and flaws, and cyber-attacks on Wi-Fi networks. The sources were selected based on a set of criteria and keywords, such as year and type of publication, and publication libraries.
In a first part each selected source is summarized, highlighting their individual focus on the discussed threats or attacks and the resulting findings. The second part, the literature-review, consists of interwove comparisons of the sources' topics and findings, categorized by threats or attack types.
The literature-review concludes that many of the presented threats and attacks are enabled by inherent vulnerabilities in Wi-Fi protocols or implementation flaws. Some vulnerabilities may have been partially addressed in amendments to Wi-Fi standards, while others persist due to backward compatibility requirements.
Regarding improvements and future fields of study, the literature-review recognized the need for more rigorously defined standards in Wi-Fi technology. Implementations should be formally verified in a way to eliminate lacking adherence to standards and to reduce risks of bugs. Testing of Wi-Fi implementations must be expanded to include a broader range of devices, real-world environments, and configurations. This includes vendor-specific features and implementations, which often rely on Wi-Fi standards but due to ambiguous specifications lack security
Konzeption und Umsetzung eines ML-basierten Recommender-Systems für Freizeitaktivitäten
No full textEinleitung: Diese Arbeit beschäftigt sich mit der Entwicklung eines personalisierten Empfehlungssystems für Freizeitaktivitäten, das psychologische Merkmale der Nutzer (Big-FiveModell) mit standortbasierten OSM-Daten kombiniert. Anders als herkömmliche Verfahren, die umfangreiche Nutzerinteraktionsdaten erfordern, stützt sich unser Ansatz auf wissenschaftlich validierte Persönlichkeitsprofile und ein kontinuierliches Like-/Dislike-Feedback zur Feinjustierung der Empfehlungen.
Vorgehen: Zunächst wird ein orthogonal gestalteter Fragebogen eingesetzt, um die anfänglichen Nutzermerkmale zu erheben und daraus einen Nutzervektor abzuleiten. Das Empfehlungssystem nutzt einen zweistufigen Algorithmus: Phase 1 (Klassenbasiert) für Kaltstart und wenige Tag-Learnings, Phase 2 (OSM-Tagbasiert) für eine feingranulare Anpassung, sobald genügend Tag-Vektoren vorliegen. Hierbei misst ein Matching-Algorithmus die euklidische Distanz zwischen Nutzer- und Aktivitätsvektoren. Die technische Umsetzung umfasst ein Flask-Backend (Datenverarbeitung und Kommunikation mit der Overpass API) und ein React Frontend (UI für Suche, Feedback, Visualisierung). Eingebundene Echtzeit-POI-Daten aus OSM werden dynamisch klassifiziert und in unsere Empfehlungslogik integriert.
Ergebnis: Die Evaluierung verdeutlicht, dass das System bereits mit begrenztem Feedback relevante und stimmige Empfehlungen erzeugt. Der zweistufige Ansatz ermöglicht robuste Ergebnisse bei Kaltstarts (Phase 1), während er bei weiterem Feedback (Phase 2) immer individuellere Vorschläge liefert. Die kontinuierliche Anpassung der Nutzerprofile über Like-/Dislike-Eingaben steigert die Personalisierung schrittweise. Erste Tests bestätigen die Lernfähigkeit und Stabilität des Systems. Zukünftige Weiterentwicklungen könnten kontextbasierte Faktoren (Wetter, Events) oder ein federiertes Lernverfahren integrieren, um Privatsphäre zu erhöhen und das Empfehlungsspektrum weiter zu verfeinern