21793 research outputs found
Sort by
Not Only the Last-Layer Features for Spurious Correlations: All Layer Deep Feature Reweighting
Get PDFSpurious correlations are a major source of errors for machine learning models, in particular
when aiming for group-level fairness. It has been recently shown that a powerful approach to
combat spurious correlations is to re-train the last layer on a balanced validation dataset, isolating
robust features for the predictor. However, key attributes can sometimes be discarded by neural
networks towards the last layer. In this work, we thus consider retraining a classifier on a set of
features derived from all layers. We utilize a recently proposed feature selection strategy to select
unbiased features from all the layers. We observe this approach gives significant improvements
in worst-group accuracy on several standard benchmarks. Another pain point in transfer learning
is with out-of-distribution tasks having large distribution shifts relative to the source task. Full
finetuning suffers in performance as it disturbs backbone parameter weights during the starting
few optimization steps and is forced to make drastic adaptations to correct for large losses initially
observed in training. Linear tuning is another approach shown to improve model generalization
capabilities and is especially effective for transfer learning on out-of-distribution downstream tasks.
We further evaluate the usefulness of intermediate layer information by incorporating it with a linear
tuning approach. Results over datasets from a common visual task adaptation benchmark show that
the empirical benefits from simply leveraging intermediate layers are similar to the proposed method
and there is no noticeable gain in accuracy from incorporating a linear tuning step
High Gain Millimeter-wave Antenna Array Design with High Isolation
Get PDFMillimetre wave (mm-wave) antenna arrays with high gain are explored to meet the user requirements of high throughput, with a very close distance between the antenna elements, which enables miniaturisation. However, high-gain antenna array design at mm-wave bands has been a growing concern among antenna engineers because of the inevitable, undesired mutual coupling between the antenna elements. First, various decoupling techniques are explored to increase isolation between patch and ME-dipole antenna arrays at 30 and 60 GHz. Then, a novel
customized π-shaped split-ring resonator (SRR) metasurface is designed. The SRR is arranged in two configurations to decouple 1×2 and 1×4 millimeter-wave (mm-wave) magneto-electric dipole (ME-dipole) in the H-plane. The antenna performances are verified. Third, an effective method is used to design a large dual-polarized finite planar array and its corporate feed network. The procedure is verified by an 8×8 and 16×16 array of metallic ME dipoles fed by a network of
Microstrip Ridge Gap Waveguide (MRGW). The procedure is based on designing the corporate feeding network by replacing the elements’ ports with each element’s corresponding effective input impedance that accounts for the mutual coupling between the antenna elements. The results are verified by the full-wave numerical solution. A 10×10 array is fabricated and measured. The array bandwidth is similar to the element bandwidth. The simulated results are confirmed by measurements. Finally, the same technique is used to design a dual circularly
polarized metallic magnetoelectric dipole with excellent radiation characteristics. Different scenarios are explored, and the antenna is fabricated to confirm the simulated results. The single-layer, single-slot design is also introduced to save the cost of fabricating dual slots. The two-layer feeding network is introduced to solve the problem of extra gaps introduced by the feeding network. The performances
are compared with the single-layer excitation
Functional End-to-End Testing of IoT Systems
Get PDFIoT systems are increasingly deployed across various domains, making end-to-end (E2E) testing crucial to ensuring expected functionality. However, testing IoT systems is challenging due to their heterogeneity, distributed execution, and real-world constraints. Traditional testing approaches are inefficient and limited, focusing on isolated layers rather than complete system interactions.
This thesis presents findings from a Systematic Literature Review (SLR) and an Industry Study on IoT system testing, analyzing existing challenges, approaches, and tools. Based on these insights, we propose a taxonomy for testing IoT systems and introduce a framework for evaluating the technical software engineering (SE) testable aspects of IoT systems. Building on this foundation, we propose an approach for functional end-to-end (E2E) testing of IoT systems, leveraging Use Case Specifications (UCSs) written in a restricted format and IoT systems descriptions. Our approach systematically converts UCSs into executable test scenarios, which are further transformed into structured test data (i.e., payload). This payload provides the necessary data for generating test cases that cover multiple layers of the IoT system. The generated test cases are then executed on the system under test (SUT) to detect bugs. To evaluate the proposed approach, we conducted an empirical study on an IoT system, analyzing its effectiveness in test case generation and bug detection. The results demonstrate that our approach detects bugs across multiple layers of the IoT system by leveraging the use of real-time execution data. Furthermore, it significantly improves test coverage and efficiency, reducing manual effort while maintaining accuracy. These findings indicate that the use of real-time execution data at each layer enhances bug detection in IoT systems
Experimental and numerical investigations on compressive and tensile responses of heterogeneous soils in cold environments
Get PDFClimate change is profoundly affecting permafrost regions, posing significant challenges to infrastructure stability. As global temperatures rise, permafrost thaws, leading to ground subsidence and compromising structures such as roads, pipelines, and buildings. Understanding soil behavior under thermo-mechanical forces is crucial for designing resilient engineering solutions. This thesis offers a comprehensive study of the thermo-mechanical behavior of cold-region soils through experimental data and numerical analyses under compression, tension, and triaxial conditions. The research investigates temperature-dependent strength variations, effective measurement techniques for the tensile behavior of frozen soil, and the rheological and residual strength characteristics of lime-treated (L-soil) and untreated natural soil (N-soil) from northern Quebec, Canada, after freeze-thaw cycles. The key contributions of this work include: (I) identifying effective and reliable techniques for quantifying tensile strength; (II) determining the critical number of freeze-thaw cycles to predict residual strength in L-soil and N-soil; (III) analyzing failure modes and stress behavior in L-soil and N-soil under varying confining stresses and thermal conditions; (IV) accurately modeling visco-elastic, visco-plastic, and creep behavior of compressive and tensile strength using finite element methods; and (V) calibrating triaxial testing approaches against experimental data. Finally, this study simulates the damage initiation and crack propagation in uniaxial compressive test and indirect tensile test using damage XFEM model in finite element-based software package (Abaqus). Findings emphasize the importance of considering time-dependent strength degradation, analyzing the complex behavior of frozen soil due to interactions between frozen and unfrozen water, and quantifying tensile strength with minimal local plastic deformation. The research also highlights the benefits, drawbacks, and challenges of using lime to enhance soil residual strength in cold climates.
Despite significant advancements, the research acknowledges limitations, such as the need for microscopic studies of unfrozen pore water and ice interactions, scanning electron microscopy (SEM) analyses of lime, silt, fine sand, and clay particle interactions, and broader validation efforts for real-world scenarios. Recommendations for future work include micro-level soil sample studies, dynamic and creep loading tests under various temperature and moisture conditions, modifying hyperbolic Drucker-Prager modeling to account for temperature-dependent parameters, conducting long-term studies, and integrating field data to enhance model accuracy and applicability
The Impact of CEO Activism on ESG Ratings and Analyst Forecast Accuracy
No full textThis dissertation examines the role of CEO activism as a form of nonfinancial communication that
shapes how firms are evaluated by external market intermediaries. As CEOs increasingly take
public stances on controversial sociopolitical and environmental issues, questions arise about how
such behavior is interpreted by third-party experts and whether it influences firm-level outcomes.
Focusing on two key information intermediaries—ESG rating agencies and sell-side financial
analysts—this dissertation explores the dual role of CEO activism: as a reputational signal of
corporate advocacy on public policy and social reform and as a potential source of value-relevant
information.
The first chapter investigates whether CEO activism is associated with higher ESG ratings. Using
hand-collected data on CEO activism events among S&P 500 firms from 2010 to 2019, the study
assesses whether public statements made by CEOs on societal issues are perceived by ESG
agencies as credible signals of corporate advocacy on contentious societal issues or dismissed as
symbolic gestures. The findings suggest that activism can positively influence ESG ratings,
particularly pronounced when CEOs speak out on more controversial issues where the credibility
of CEO activism gesture is heightened.
The second chapter explores how financial analysts respond to CEO activism by examining its
effect on analyst forecast accuracy. Analysts, as forward-looking information intermediaries, may
interpret CEO activism as a useful and credible nonfinancial information that signals strategic
direction and corporate value—or as low-quality noise that increases uncertainty. The results show
that, on average, CEO activism improves forecast accuracy, especially in firms with stronger
governance, greater transparency, and less controversial issue engagement.
Collectively, the findings contribute to the literature by framing CEO activism as both a genuine
expression of corporate values and a useful form of nonfinancial communication. In ESG
evaluations, CEO activism is seen as a credible expression of corporate leadership and linked to
higher ESG ratings. Analysts, by contrast, treat it as a source of nonfinancial information that
informs forecasts. Together, the chapters show that the impact of CEO activism depends on the
lens of the external experts, shaping how executive voice is understood in capital markets
The Merging Sounds of Bells & Electronic Dance Music: Notes on Taiwanese Expressions of National Identity in the Dajia Mazu Pilgrimage
Get PDFThe Dajia Mazu pilgrimage, organized annually by the Zhenlan Temple in Taichung, is considered the most important pilgrimage in Taiwan. Each year, more than a million people participate in this event, desiring to walk alongside the chariot carrying the temple’s central statues of Mazu. Although religious, it is not unusual to find during the event examples of spectacles which are in discordance with preconceived ideas of religious sanctity such as the performance of Electronic Dance Music, expositions of modified automobiles, demonstrations of beauty contests, and the presence of stripper dancers. These non-religious productions imply a complex interaction between cultural expressions of Taiwanese identity and an attachment to religious values. Using ethnographic data; personal observation; and relying on previous scholarship; this thesis portrays the Dajia Mazu pilgrimage as a platform on which the Taiwanese followers of Mazu can express their cultural identity and uniqueness in a response to the growing political tension with mainland China. By looking at the religious significance of the festival, the cultural performances presented during the pilgrimage, and the political intervention in its promotion and organisation, this work highlights how the Taiwanese utilize the worship of Mazu as a vehicle to convey an independent cultural identity. Using the Dajia Mazu pilgrimage as a case study, this research contributes to the body of scholarship discussing Taiwanese’s religions and culture, national identity, and geo-political relationship with China, while simultaneously addressing disregarded aspects of the pilgrimage such as religious and cultural performances, the development of a material culture, and more
Robust Design of a Manufacturing Network for Mass Personalization
Get PDFThe Fourth Industrial Revolution (Industry 4.0 or I4.0) is transforming manufacturing through the integration of cyber-physical systems, artificial intelligence, and the Internet of Things. At its core is mass personalization (MP), enabling the production of customized products, particularly in high-tech sectors such as aerospace, medical devices, and precision optics. These industries require resilient supply networks to handle low-volume, high-complexity production and uncertainties in customer demands and supplier performance. Traditional supply chain models fall short in addressing these challenges, calling for advanced optimization frameworks.
This thesis explores the design of resilient and reconfigurable supply networks tailored to MP under I4.0. It makes three primary contributions. First, a strategic mixed-integer programming (MIP) model is proposed for optimizing supplier selection and order allocation, balancing design complexity with economies of scale. Second, a two-stage stochastic programming (2SP) model is developed for platform-based manufacturing networks, integrating crowdsourcing to enhance resilience by assigning primary and backup suppliers under uncertain capabilities. Third, an adjustable robust optimization (ARO) model is introduced for multi-echelon networks, addressing variability in supplier capacity and bill-of-material complexity, supported by an efficient math-heuristic algorithm.
Extensive numerical experiments and sensitivity analyses validate the models’ effectiveness in mitigating risk and improving resilience. This research offers actionable insights for high-tech manufacturers aiming to build agile, cost-efficient supply networks that meet the evolving demands of mass personalization in the era of Industry 4.0
Detection of Dangerous Driving Behaviors Using Multi-Dimensional Data-Driven Methodology
No full textTransportation technologies are currently experiencing rapid advancements in the context of global development. The rapid increases higher in traffic volume, however it has contributed to higher, and severity of traffic are continuously rising. The preservation of human lives and attributed to road traffic accident has become vital worldwide and the public is an urgent requirement. However, owing to frequent and widely diverse dangerous driving behaviors of drivers, there are significant potential road safety risks, which will seriously affect the healthy development in the transportation industry. To mitigate such safety hazards, there is an urgent need for precise detection and warning of dangerous driving behaviors of human drivers under driving condition. This dissertation research focuses on a multi-scale data-driven method for detecting drivers’ dangerous driving behaviors. The multi-dimensional data basis provided the essential providing theoretical basis and technical support platform is proposed for formulating of a comprehensive driver warning system.
The main work of this paper includes the following five parts:
(1) Analyzing the danger of a driving behavior and its relationship with traffic accidents from the perspectives of public safety and system engineering is developed considering a visual recognition detection system architecture for dangerous driving behaviors based on using the reported research methods, and deep learning neural network models and detection algorithms used for dangerous driving behavior recognition, laying the theoretical foundation for subsequent research.
(2) An efficient identification method is subsequently proposed for dangerous driving behaviors based on the improved YOLOv8 (You Look Only Once version 8) neural network platform. To improve the recognition efficiency and accuracy of dangerous driving behavior detection, a Multi-Head Self-Attention (MHSA) attention mechanism module is further adopted to enhance efficiency and accuracy, enabling the model focus on global target within a larger receptive field, thereby enhancing the recognition capability of targets. This global modeling capability helps reduce false positive and false negative rates in target detection, improving the overall performance and robustness of the model. Meanwhile, inserting a driver's driving emotion detection module in the network layer shares ROI (Region Of Interest) features in the target detection algorithm, effectively increasing the detection of driver's driving emotion recognition function without significantly increasing the complexity of network computation demand.
(3) Proposing a method for detecting driver's brake pedal and accelerator pedal operations using the Mask-RCNN instance segmentation deep learning network. It recognizes and evaluates the operation of driver's brake pedal and accelerator pedal through video frames collected using a driver's leg camera. The use of ROI Align method to effectively align pixels during downsampling enhances the feature extraction capability of the entire detection network with increasing the computational efficiency and complexity of the network model, thereby improving the efficiency and accuracy of the detecting driver's brake pedal and accelerator pedal operations.
(4) A data-driven detection method is subsequently proposed using the filtering and sliding windows. By collecting brake signals and throttle signals from the vehicle CAN bus data and performing low-pass and median filtering, the computational feature extraction capability of the driving signal data is enhanced. The sliding window method compares and judges brake signals and throttle signals in the sampling interval with the corresponding thresholds, achieving detection and evaluation of driver's dangerous driving behaviors such as sudden acceleration and deceleration.
(5) Building a dangerous driving behavior detection system, integrating neural network training weights and detection algorithms through Python and Qt Designer software systems, and designing a visual real-time detection front-end UI interface. At the same time, designing real vehicle verification experiments, based on real-time experiments in actual driving environments, to verify the effectiveness and superiority of the driver's dangerous driving behavior detection method proposed in this study, providing theoretical basis and technical support for the widespread application of dangerous driving behavior recognition technology
Reengineering MSCI’s ESG Ratings Methodology
No full textThis study builds on the existing Morgan Stanley Capital International (MSCI) ESG Ratings methodology and introduces a complementary framework designed to better incorporate supply chain considerations and dynamic risk factors. While MSCI’s established model effectively evaluates Environmental, Social, and Governance (ESG) factors using fixed weights and proportional adjustments, there is an opportunity to further refine the approach by incorporating industry-specific key issues, regional ESG risk variations, and the interconnected nature of ESG dimensions. To support this, we propose enhancements across seven key areas, such as dynamic redistribution of key issue weights, governance scoring that reflects industry context, and controversy assessments that factor in a company’s remediation efforts.
A key contribution of this work is the introduction of a dedicated Supply Chain pillar, transforming ESG into ESSG (Environmental, Social, Supply Chain, and Governance). This expansion
addresses the systemic oversight of upstream risks and labor violations, which often go unnoticed under the current three-pillar model. We argue that supply chain transparency, ethical sourcing,
and logistical resilience should be evaluated independently, given their growing relevance to financial, operational, and reputational performance of companies.
Furthermore, we propose a dynamic weighting framework that adjusts key issue scores based on their relative importance within an industry context. Unlike the current approach, which
redistributes weights uniformly, our model assigns weights based on issue criticality.
We further address the limitations of static geographic and governance weightings by introducing context-sensitive exposure and management adjustments. These include a flexible geographic risk multiplier and flexible governance weighting based on how strict the rules are in different industries. Controversy deductions are also restructured using a remediation scoring mechanism,
enabling companies to partially offset penalties by demonstrating corrective actions.
Finally, we propose for an integrated approach to ESG scoring that captures interdependencies across pillars, recognizing that risks and opportunities often intersect. For instance, poor social
practices in supply chains can intensify environmental harm, a connection the current siloed methodology often fails to capture.
Through these enhancements, our proposed framework delivers a transparent, and actionable assessment of corporate sustainability. It encourages continuous improvement, supports long-term
planning, and better aligns ESG ratings with the realities of globalized business operations.
Keywords: ESG Ratings, Supply Chain Sustainability, ESSG Framework, Remediation Scoring, Dynamic Weighting, Industry-Specific Risk Assessment, Cross-Pillar Interdependencies, ESG
Exposure Adjustment
Negative Utopia: The Configuration of a Marxian Method of Speculation
No full textKarl Marx is often considered to be an anti-utopian thinker on the grounds of his and Friedrich Engels’s critique of the “utopian socialists” in the Communist Manifesto. However, Marx’s opposition to these early socialist thinkers stems from the lack of utopian aspiration that their projections profess. For Marx, the “phalanesters” that these thinkers envision are constrained by bourgeois aspirations; rather than putting forward effective transformative plans, they reinforce reactionary social arrangements. In an effort to reaffirm the utopian valence of Marx’s critique, this thesis presents a re-reading of Capital Volume 1 to consider the formulation of the “negation of the negation” as a methodological proposition of utopian character. By defining the utopian essence of Marx’s dialectical critique as negative utopia, I argue that, in negating the absolute character of capitalist forms of appearance, Marx uncovers the postcapitalist possibilities that exist within capitalist mechanisms of abstraction, equation, and alienation.
To underscore the utopian character of Marx’s critical method, this thesis also traces the development of this form of analysis, of a negative utopia, in the works of Marxist thinkers, Georg Lukács, Theodor Adorno, Ernst Bloch, Herbert Marcuse, and Fredric Jameson. The genealogy presented here demonstrates the prevalence of this utopian impulse in the progression of Marxist theory, particularly in works which consider ideological critique as central to Marxian thought. This review of the ideas of these thinkers highlights the developments and expansions they introduced, incorporating elements of non-identity, desire, and consciousness into the conception of utopia, thereby enriching both the speculative and utopian dimensions