21793 research outputs found
Sort by
Retrieval Augmented Chatbots powered by Large Language Models for Semantically Structured Data
Get PDFRecent advancements in Large Language Models (LLMs) have transformed Natural Language Processing, yet challenges such as factual inaccuracies and inadequate reasoning over structured data persist. Retrieval-Augmented Generation (RAG) systems address these issues by grounding LLMs in external knowledge. However, conventional RAG methods typically treat knowledge sources as unstructured text, overlooking the semantic relationships vital in domains like enterprise data and healthcare. This thesis introduces a Graph-based RAG approach that leverages the structured nature of graph data to enhance both retrieval and response generation by preserving these semantic relationships. The research focuses on developing conversational question answering systems over semantically structured data, specifically targeting JIRA Issues and Knowledge Graphs through two distinct applications. The core innovation lies in maintaining the inherent data relationships during both retrieval and generation phases by employing structured queries and graph traversal techniques. This method not only allows for domain-specific optimization but also demonstrates improved performance and efficiency compared to traditional RAG methods
XFEM based Multiscale Approach for the Analysis of Masonry Walls
Get PDFThe increasing global emphasis on sustainable development has underscored the importance of understanding the mechanical behaviour of masonry materials in the context of modern construction practices. This thesis presents an X-FEM-based computational homogenization framework tailored for analysing heterogeneous masonry structures. The methodology leverages the advantages of the eXtended Finite Element Method (X-FEM) in introducing phased changes and capture interface behaviour within the representative volume element (RVEs) to represent the material behaviour of masonry structures.
The proposed framework introduces techniques for calculating effective material properties, incorporating periodicity of the masonry wall structure, and addressing interface damage effects. A detailed derivation of the governing equations is presented, along with rigorous validation against benchmark studies and experimental results from the literature. The analysis encompasses multiple case studies, including parametric studies on RVE size, assumed boundary condition effects, and the impact of interface damage on homogenized properties. Results reveal the accuracy and robustness of the X-FEM based approach in capturing the homogenized behaviour masonry walls.
This study not only bridges gaps in existing computational techniques but also provides insights into optimizing modelling strategies for masonry homogenization. The developed framework holds significant potential for applications in structural analysis and the sustainable design of masonry structures, offering a reliable tool for engineers and researchers in the field of computational mechanics
Why Relational Facts need to be at least as Primitive as Monadic Facts
No full textThe problem of relations has turned on the question of whether or not relations are real “things” (in the same sense as “objects are things”). This is a mistake and has been the reason so little has been accomplished in the discourse about how to deal with relational sentences and relational facts. I claim instead we should be concerned with the facts which make relational sentences true, and whether those facts are primitively relational, insofar as they cannot be analysed into non-relational, monadic facts. With this claim defended, I argue that it becomes clear that some facts must be primitive relational facts
An AI-Powered Framework for Processing and Analyzing Climate Action Plans
Get PDFClimate action plans play a critical role in global efforts to combat climate change. They outline strategies for reducing greenhouse gas emissions and enhancing climate resilience. However, these plans exist in diverse, unstructured formats, making automated extraction, comparison, and analysis challenging.
This thesis presents an AI-powered framework that leverages natural language processing (NLP) and machine learning techniques to process climate action plans systematically. The
framework performs document preprocessing, text chunking, and information extraction to convert unstructured reports into structured, queryable data. The extracted content is stored in two
distinct databases:
(1) An action plan database, which captures detailed policy strategies, governance mechanisms,
and implementation stages.
(2) A progress reports database, which enables tracking of policy evolution over time.
The framework facilitates structured querying and analysis, allowing stakeholders to compare climate strategies across different regions and timeframes. A key feature is the progress report generator, which systematically identifies added, removed, and modified actions between different versions of a plan, providing insights into policy effectiveness. Experimental results demonstrate the framework’s ability to process diverse climate action plans, extract structured data, and generate insights into trends, policy focus areas, and implementation progress. By offering a scalable, data-driven approach to climate policy analysis, this research contributes to the field of environmental data science and supports evidence-based decision-making for climate action
A Safety-Focused Systems Architecting Framework for Aircraft Conceptual Design
Get PDFTo reduce the environmental impact of aviation, aircraft manufacturers develop novel aircraft configurations and investigate advanced systems technologies. These new technologies are complex and characterized by electrical or hybrid-electric propulsion systems. Ensuring that these complex architectures are safe is paramount to enabling the certification and entry into service of new aircraft concepts. Emerging techniques in systems architecting, such as using model-based systems engineering (MBSE), help deal with such complexity. However, MBSE techniques are currently not integrated with the overall aircraft conceptual design using automated multidisciplinary design analysis and optimization (MDAO) techniques. Current MDAO frameworks do not incorporate the various aspects of system safety assessment. The industry is increasingly interested in Model-Based Safety Assessment (MBSA) to improve the safety assessment process and give the safety engineer detailed insight into the failure characteristics of system components early in the design process. This thesis presents a comprehensive framework to introduce aspects of the SAE ARP4761 safety assessment process in conceptual design while also considering downstream compatibility of the system architecting and safety assessment processes. A generic element architecture description approach, implemented using a graph-based system architecture descriptor, is introduced to model and transfer system architecture information between each stage of the systems architecting process while supporting safety assessment activities at multiple levels of architecture granularity. The proposed framework introduces a safety-based filtering approach for large system architecture design spaces and integrates quantitative safety assessment methods compatible with early-stage system architecture specifications. Furthermore, the generic element descriptor links early system architecture specification with formal architecture specification in an MBSE environment. The framework also enables both simple and formal system architecture specification models to be used as inputs to safety assessment, as well as a source of system-level sizing parameters for MDAO workflows featuring system sizing tools. The framework’s effectiveness is illustrated with examples from applications in recent collaborative research projects with industry and academia, which feature safety-focused system architecting studies for a conventional aircraft landing gear braking system, a yaw control actuation system, and unconventional yaw control system architectures for hybrid-electric and distributed-electric aircraft. The work presented in this thesis contributes to increasing maturity in conceptual design studies and fosters innovation by opening the design space while considering safety upfront
securing 5g o-ran systems against resource depletion and synchronization attacks
No full textThe Open Radio Access Networks movement (O-RAN) drives the industry toward open interfaces such as the open fronthaul (O-FH) and enables RAN virtualization and intelligence through AI by augmenting the RAN with radio intelligent controllers capable of hosting various applications that optimize the RAN’s performance and potentially bolster its security.
However, this openness and intelligence in design expand the attack surface and present new challenges in ensuring network security. In this thesis we present two contributions aimed at addressing security challenges both inherited from previous RANs or specific to the O-RAN.
We address the Base Transceiver Station Resource Depletion (BTS-RD) attack, an attack that exploits the lack of integrity check in the Radio Resource Control protocol to deplete the BTS resources. However, in this work, we leverage O-RAN intelligence to introduce a novel O-RAN-compliant detection solution that detects variations of the BTS-RD attack. Our solution, the “BTS-Band”, employs a Bidirectional Long Short-Term Memory Autoencoder (BiLSTM-AE) for detecting anomalies. Moreover, to promote trust we integrate Shapley additive explanations to provide explanations of the model’s decisions. To evaluate the performance of the BTS-Band we built a 5G testbed based on OpenAirInterface, on top of which, we emulate and extract benign and attack traffic to train and test our solution. Accordingly, the BTS-Band achieves an average F1-score of 92.4% in detecting multiple BTS-RD variations.
Second, we shift the focus to our work on assessing the security of the O-FH. Particularly, the security of the newly introduced Synchronization Plane (S-plane) is interesting for its sensitivity to latencies and performance expectations that makes integrating appropriate security measures challenging. We identify emerging security threats on the S-plane, discuss their impact on O-RAN synchronization topologies, and accordingly present countermeasures.
As such, our work contributes to enhancing the resilience and security of next-generation open RAN systems
Does Intranasal Oxytocin Reduce Symptoms of Mental Disorders? A Meta-Analysis of Clinical Trials
No full textIntranasal administration of oxytocin has been shown to enhance social cognition and reduce stress reactivity in healthy individuals, indicating potential therapeutic benefits for mental disorders. However, clinical trials have produced mixed results. Following a systematic search, data were extracted from 42 double-blind, randomized controlled trials comparing symptoms following intranasal oxytocin versus placebo in autism spectrum disorder, schizophrenia spectrum disorders, substance use disorder, and other mental disorders. Random effects meta-analysis of the pooled sample (N = 2185) revealed a small, non-significant overall treatment effect with substantial between-trial heterogeneity (g = 0.17, 95% CI = –0.02 to 0.36, I² = 77.41%). The removal of two outlier studies with extremely large treatment effects in substance use disorder caused a significant moderation by mental disorder category to disappear. However, the removal of these outliers also revealed a significant moderation by biological sex whereby studies with more females showed greater treatment effects. No significant moderation by dose, number of administrations, or psychosocial interventions was detected. Despite promising findings in individual studies, intranasal oxytocin is not currently supported as an evidence-based treatment for mental disorders. Future clinical trials should systematically examine dose-response relations, optimize psychosocial intervention protocols, address the underrepresentation of females, and report individual participant data which will enable meta-analyses to investigate individual differences in treatment response
How different types of brand purpose ads effect consumer attitude towards the ads: The mediation effect of inspiration and downstream consequence on consumer behavior.
No full textAccording to advertising research, brand purpose advertisements can be classified into three types: those that use positive emotional appeals, negative emotional appeals, or a mix of both. While much of the earlier literature has primarily concentrated on either positive or negative appeals, the concept of mixed emotional appeals is a newer area of interest. This study is the first to further divide mixed brand purpose ads into two categories: concrete mixed appeals and abstract mixed appeals. The primary objective of this research is to explore how abstract and concrete mixed emotional appeals affect consumer attitudes toward advertisements. It also examines the mediating role of inspiration, the moderating role of self-construal, as well as the downstream consequence on behavioral intentions. Through two experimental studies, the findings reveal that abstract mixed appeals lead to more favorable consumer attitude towards the ad and this effect is mediated through inspiration. Furthermore, findings show that favorable attitude towards the ad leads to behavioral intentions
AI Driven Transformation of Building Assessment Report into Energy Models for Building Portfolio Analysis
No full textBuilding energy modeling is essential for achieving energy optimization and net-zero targets, yet portfolio-scale analysis faces significant constraints due to critical building information remaining trapped within unstructured documentation such as permits, energy audits, and maintenance records that cannot be systematically extracted and analyzed. This research presents an AI-driven framework utilizing local Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) to automatically extract and structure building information while ensuring complete data privacy and security.
The methodology employs on-premises LLM deployment to process sensitive documentation without external data transmission. This approach efficiently transforms diverse unstructured building data into standardized inputs for the Honeybee Python modeling platform, automatically generating detailed energy models and producing comprehensive portfolio-wide performance analytics. The resulting portfolio-wide analytics identify key patterns, inefficiencies, and optimization opportunities across building inventories, while the detailed energy models provide enhanced baseline representations that can be further refined by energy modelers or improved through more sophisticated RAG implementations for deeper analysis.
Validation across 80 Quebec office buildings demonstrated 92.5% processing reliability while reducing analysis timeframes from days or weeks to just a few hours—representing substantial improvements in portfolio assessment efficiency. The framework successfully addresses persistent data processing limitations that have constrained evidence-based energy management, enabling the systematic, data-driven portfolio strategies necessary for achieving ambitious decarbonization objectives
Autonomous Real-time Forest Fire Detection and Firefighting using Unmanned Aerial Vehicles
No full textThis thesis proposes a framework for autonomous real-time fire detection and firefighting using Unmanned Aerial Vehicles (UAVs). In recent years, wildfires have inflicted increasingly severe damages on ecosystems and human settlements. UAVs have demonstrated strong effectiveness in forest fire detection as they provide multiple onboard sensors and can rapidly cover large, inaccessible areas. Forest firefighting using UAVs remains an emerging field of study, and although UAVs offer significant advantages for forest firefighting due to their mobility, flexibility, and ability to operate in hazardous environments, developing reliable algorithms for this application presents several challenges. In fact, as the amount of payload that can be carried by a UAV is limited, water or fire retardants must be dispersed precisely. The importance of accurate water dropping will be highlighted further by considering that the UAV's onboard GPS sensor also introduces errors to the system. Moreover, a robust UAV wildfire suppression for a line of fire scenario, which is the most common form of forest fire occurrence, is not presented in the existing literature. Hence, this thesis proposes a framework in which a UAV first runs a forest fire detection model to find the direction of fire, followed by initiating a particular UAV pattern optimized for a fire spot localization algorithm, and then upon acquiring the GPS coordinates of the fire spots, the UAV suppresses the line of fires, all within a single flight. A visual feedback is integrated with a designed control system, which adjusts the UAV to the desired position prior to the release of the fire retardant. The developed algorithms are verified through experimental outdoor flight tests using a DJI M300 RTK UAV with an H20T camera system and a designed water tanker system to be mounted on the UAV