21793 research outputs found
Sort by
Automated Window-to-Wall Ratio Estimation Using Google 3D Map Tiles Analysis and AI
No full textThe Window-to-Wall Ratio (WWR) is a critical parameter in architecture and urban planning, influencing energy efficiency, daylighting, and thermal performance. Traditionally, WWR is either manually measured or obtained from architectural plans, but these methods are impractical for large-scale studies due to data unavailability. Recent advances in machine learning and 3D mapping offer new opportunities for automation. In this work, we propose an automated method to estimate WWR by extracting 3D building meshes from Google 3D Maps and using a machine learning model to segment windows and walls from their images. We utilize Google 3D Maps for 3D building model extraction and images, OpenStreetMap (OSM) for defining building boundaries, and a UNet-based deep learning model to segment windows, walls, roofs, and other elements. The UNet model, trained on 1,000 manually labeled building images, accurately segments windows and walls, from which WWR is calculated. Unlike previous methods relying on facade images, drone surveys, or street-level views, our approach leverages Google’s pre-existing 3D city data, available for over 2,500 cities worldwide, offering a scalable, effective, and cost-efficient solution. The AI-predicted WWR values are validated by comparing them with manual measurements from actual buildings, confirming the method’s accuracy. This research advances the automation of architectural and energy analysis, providing a fast, cost-effective, machine learning-driven tool for WWR estimation. Future work includes enhancing model robustness, incorporating larger datasets, and exploring additional 3D data sources
The Soldier’s Dilemma - To Fight or Not to Fight: A Study of World War II Defection Through Soviet Leaflets Targeting Wehrmacht Soldiers
Get PDFWhen in 1941 Nazi Germany invaded the USSR, Soviet authorities, aware of the Wehrmacht’s initial superiority, hoped they could weaken the enemy by convincing some of their soldiers to defect and surrender. That idea resulted in a large propaganda campaign, whereby the frontlines were flooded with propaganda leaflets, most notably the series titled the ‘Front-Illustrierte für den Deutschen Soldaten.’ For the Soviets, however, defection was not intended to provide a chance for German soldiers to fight alongside the Red Army. Instead, as will be outlined in my analysis of the leaflets in this collection, the Soviets hoped to convince the German forces that Hitler had betrayed them and Germany, and that the best thing they could do to protect their homeland was to voluntarily surrender to the Red Army. To convey these ideas, Soviet propagandists sought to open a rift between the members of the Wehrmacht and the Nazis by pointing out the multiple grievances German soldiers should have had against the Nazi government. Simultaneously, soldiers were told that they could spend the rest of the war, if only they surrendered, in a cozy prisoner of war camp where they might engage in an array of leisure activities, which included everything from playing games to reading books, and partake of services that they would have enjoyed while back home in Germany, such as haircuts and warm baths. The Soviets then hoped that, combined, the grievances pointed out to German soldiers and the appealing alternatives they were offered might convince enough of them to defect and surrender, and that the outcome of the war might be affected positively
When the Maiden Becomes Death: Contemplating the Monstrous Girl in the Fairy Tale Reimaginings of Angela Carter and Helen Oyeyemi
Get PDFWhen Helen Oyeyemi’s wicked stepmother, Boy, shows her friend her nightmarish engagement gift from her husband-to-be—a handcrafted bracelet of a white-gold snake whose body wraps around the length of Boy’s forearm—her friend replies: “I mean, could that scream ‘wicked stepmother’ any louder?” (110). Boy’s bracelet foreshadows her fate as the wicked stepmother, but her evilness is equivocal: she exiles one daughter to protect another. Angela Carter’s Snow White seems to become somewhat monstrous, too, after she is violently assaulted. When Angela Carter and Helen Oyeyemi retell a fairy tale, they reject the black-and-white lens that the conventional fairy tale is told through. Both writers unsettle the fairy tale tradition by unearthing and exposing the darker meanings that remain buried in these original tales. I argue that it is most notably through the figure of the monstrous girl that Carter and Oyeyemi foreground the patriarchal and misogynist politics that the fairy tale encodes yet overlooks. Carter and Oyeyemi, in their respective “Snow White” and “Sleeping Beauty” retellings, turn the familiar character of the innocent girl monstrous to emphasize what it means to be monstrous—and that to be monstrous is not a simple question of being virtuous or cruel, good or bad. Carter’s and Oyeyemi’s female monsters are complex, and I further examine how their actions originate from a desire to attain agency over their bodies. When Carter and Oyeyemi retell a fairy tale, they do not complicate the fairy tale but rather reveal that it is already intricate, filled with hidden messages
Time Out of Joint: Hauntology, Nostalgia, and Inter-Artistic Longing in Joanna Newsom's The Milk-Eyed Mender and Frank Ocean's Blonde
Get PDFThis thesis explores the spectral presence of the past in contemporary music through the lens of hauntology, focusing on Joanna Newsom's The Milk-Eyed Mender (2004) and Frank Ocean's Blonde (2016). While contemporary music is often shaped by nostalgia, hauntology offers a nuanced framework—one that captures the past's stubborn refusal to fully disappear, disrupting linear time and unsettling our sense of the present. Drawing on Jacques Derrida and Mark Fisher, I explore how these albums engage with hauntological aesthetics not only through sonic elements but also through their lyrics. While Fisher primarily developed hauntology in music as a sonic concept, I argue that lyrical elements can equally contribute to its effect. What this thesis specifically suggests is that inter-artistic longing—tensions between different artistic mediums within the lyrics—creates a hauntological quality. In Newsom's work, literary intertextuality generates a tension between literature and music, while Ocean's fragmented, cinematic narratives blur the boundaries between music and film. By analyzing these inter-artistic tensions, this thesis demonstrates how Newsom's and Ocean's lyricism evokes a spectral past that remains both familiar and elusive. Situating these albums within broader cultural and political contexts, I demonstrate how hauntology in lyric-driven music reflects unresolved personal and collective histories, as well as shifting cultural landscapes in the United States from the early 2000s to the mid-2010s
Addressing Data Scarcity with 2D Projection-Based 3D Point Cloud Semantic Segmentation
No full textDeep learning-based 3D semantic segmentation requires a large volume of training data to achieve robust generalization and prevent overfitting. However, acquiring extensive 3D datasets remains a significant challenge. This study addresses data scarcity in 3D semantic segmentation by leveraging a 2D projection-based approach, utilizing advancements in 2D segmentation techniques to enhance 3D segmentation performance. The key insight is that while 3D scene availability may be limited, an unlimited number of 2D projections can be generated, which can then be reprojected back into 3D space. This approach enables models to benefit from state-of-the-art 2D segmentation techniques while mitigating information loss during dimensional transformation. This study evaluates widely used 2D projection techniques, including spherical projection, bird’s-eye view, and perspective projection, to convert 3D data into the 2D domain. 2D CNN-based segmentation models, such as U-Net and SegFormer, are then applied to generate predictions, which are subsequently reprojected into the 3D domain for final segmentation. Experimental results demonstrate that using data from only a single area in the S3DIS dataset, our 2D projection-based method achieves a 3D IoU score of 47.21%. Moreover, when using a 1:1 train-test ratio on similar rooms, 3D IoU of 56.23% is achieved, highlighting that even with limited data, projection-based approaches offer a viable and effective solution for 3D point cloud semantic segmentation
Evaluating the Effectiveness of Large Language Models in Human Computer Online Negotiation
No full textABSTRACT
Evaluating the Effectiveness of Large Language Models in Human Computer Online Negotiation
Jai Priya Verma
Large Language Models (LLMs) are changing how automated systems engage with people during decision-making processes as they become increasingly integrated into online human
computer discussions. By analyzing their capacity to understand human intent, produce strong arguments, adjust to changing tactics, and produce positive results, this study assesses how
effective LLMs are in negotiating situations. Strategic rigidity, contextual misinterpretation, and ethical issues about bias and justice are some of the challenges that LLMs face despite their
strengths in language fluency, contextual reasoning, and data-driven decision-making. This study compares LLM-driven and human-driven negotiations to identify important aspects that
affect negotiating effectiveness, examine the shortcomings of existing models, and point up areas for improvement in LLM-driven bargaining systems. The research advances the field of
artificial intelligence's involvement in human-computer cooperation, negotiation automation, and the creation of more capable and flexible negotiation agents.
In various disciplines, LLMs have proven to be remarkably adept at producing intelligible writing, comprehending human intent, and supporting complicated problem-solving. Ongoing study is necessary to address objective issues like tempting offers, monetary factors, generating offers, and subjective aspects like satisfaction, fairness, and intent to return. Examining LLMs can improve their effectiveness, flexibility, and equity, increasing their
dependability for contract analysis, automated negotiations, and real-time decision assistance. In order to ensure that LLM based negotiation agent systems continue to be reliable and
efficient, my research is also essential for enhancing model transparency and resilience against hostile challenges. Understanding LLMs' advantages and disadvantages will help determine how responsibly they may be used in various situations, spurring advancements in AI-assisted communication and negotiation tools.
Keywords: Large Language Models, Online Negotiation, Human-Computer Negotiation, Software Studies, Artificial Intelligenc
Don’t Put all your NGOs in the Same Basket: Investigating the Role of NGOs in Implementing Reintegration Policy for Older Adults in Conflict with the Law
Get PDFWestern industrialized countries are experiencing an increase in the portion of their populations which are 65 and older. One result of this demographic shift, which is still occurring, is the increasing number of individuals in prisons and jails who are older adults. In Canada, up to 25 % of the federal prison population is considered older. Current studies focus on prison infrastructures and routines which are inadequate for older adults, with many authors suggesting that community placements are more appropriate for this population. Studies on prisoner re-entry focus on the success and use of programs for employment and substance use disorders.
This study focuses on an understudied group: older adults under carceral supervision who are provided reintegration services in the community. Grounded in the social construction of target populations framework – alongside NGO research on social constructions, advocacy and resource dependency – this project asks: in a context of the devolution of reintegration policy, what explains how re-entry NGOs act to broaden the bounds of reintegration for older adults? This research uses a four-case comparison of NGOs in Montréal, Toronto, San Francisco and Houston to assess how social constructions of older adults, combined with NGOs’ strategic action through advocacy and resource diversification impacts the bounds of reintegration policy. Newspaper thematic analysis, statistical analysis of funding streams and semi-structured interviews were conducted to compare reintegration services across NGOs. Compared to the two American NGOs, Canadian NGOs provided a greater number and range of services although there was important similarity between the Toronto and San Francisco cases. Within countries the more liberal-leaning cities, with the higher frequency of positive portrayals of aging were associated with more generous bounds of reintegration policy. The findings of this project highlight the importance of considering NGO service providers as agentic rather than passive recipients of correctional department funding. NGOs deploy social constructions, strategic funding searches and advocate to achieve their organizational goals
Towards Practical Second-Order Optimizers in Deep Learning: Insights from Fisher Information Analysis
Get PDFFirst-order optimization methods remain the standard for training deep neural networks (DNNs). Optimizers like Adam incorporate limited curvature information by preconditioning the stochastic gradient with a diagonal matrix. Despite the widespread adoption of first-order methods, second-order optimization algorithms often exhibit superior convergence compared to methods like Adam and SGD. However, their practicality in training DNNs is still limited by a significantly higher per-iteration computational cost compared to first-order methods. In this thesis, we present AdaFisher, a novel adaptive second-order optimizer that leverages a diagonal block-Kronecker approximation of the Fisher information matrix to adaptively precondition gradients. AdaFisher aims to bridge the gap between the improved convergence and generalization of second-order methods and the computational efficiency needed for training DNNs. Despite the traditionally slower speed of second-order optimizers, AdaFisher is effective for tasks such as image classification and language modeling, exhibiting remarkable stability and robustness during hyperparameter tuning. We demonstrate that AdaFisher outperforms state-of-the-art optimizers in both accuracy and convergence speed. The
Code is available from https://github.com/AtlasAnalyticsLab/AdaFisher
Integrating a Child-Centered Play Therapy Approach to Music Therapy Improvisation Facilitation for Children with Developmental Needs Within Mental Health Care Settings: A Philosophical Inquiry
Get PDFThis philosophical inquiry explored the integration of a child-centered play therapy ap-proach in music therapy improvisation for children with developmental needs within mental health care settings. Drawing on Wigram’s (2004) five key components of the therapeutic process and on Landreth (2023)’s nine principles of child-centered play therapy, emphasis was placed on how these frameworks can be connected and applied to music therapy improvisation, following procedures typical in philosophical inquiry, such as defining terms, relating ideas, and connecting diverse conceptual and theoretical systems (Aigen, 2005). Relevant literature on the developmen-tal needs of children in mental health care, music therapy improvisation with children, and child-centered play therapy was reviewed to establish a foundation for the inquiry. The analysis high-lighted key implications for addressing developmental needs through music therapy improvisa-tion, including adopting an experience-oriented approach, practicing cultural humility, considering individual developmental stages, providing trauma-informed care, and creating space for the ex-pression of physical aggression. The affordances, challenges and limitations associated with the use of music therapy improvisation realized within a child-centered approach are discussed, alongside the limitations of the study and future research directions
Faculty Member Descriptions of Universal Design for Learning (UDL) in Higher Education Online
Get PDFDespite growing emphasis on inclusion in higher education online, research on faculty awareness, implementation, and support for Universal Design for Learning (UDL) remains limited.
This study used qualitative data from a focus group with three faculty members in higher education who taught online to address the following research questions:
1. How do faculty members in higher education describe Universal Design for Learning (UDL) in courses online?
2. How do faculty members in higher education describe adding UDL in their courses online?
3. What challenges do faculty members in higher education encounter when it comes to UDL?
4. How do faculty members in higher education describe UDL-related resources?
The literature review explored key concepts and past studies on online learning formats and the common technological challenges faced by faculty. It also examined Universal Design for Learning (UDL), developed by CAST, which includes the principles of Engagement, Representation, and Action & Expression for accessible, inclusive learning. Key terms explored included Equity, Diversity, and Inclusion (EDI), Self-Determination Theory (SDT), and Web Content Accessibility Guidelines (WCAG) to fill a gap in the literature.
Findings showed that consistent with existent research, faculty members apply UDL strategies, especially Action & Expression, without formal training or awareness. Participants expressed a need for clearer institutional guidance and discipline-specific support to implement UDL more effectively. The study resulted in a proposed model for UDL support at the macro and micro levels using EDI, SDT, and WCAG in tandem with UDL