154092 research outputs found
Sort by
Depth-Bounds for Neural Networks via the Braid Arrangement
Get PDFWe contribute towards resolving the open question of how many hidden layers are required in ReLU networks for exactly representing all continuous and piecewise linear functions on . While the question has been resolved in special cases, the best known lower bound in general is still 2. We focus on neural networks that are compatible with certain polyhedral complexes, more precisely with the braid fan. For such neural networks, we prove a non-constant lower bound of hidden layers required to exactly represent the maximum of numbers. Additionally, under our assumption, we provide a combinatorial proof that 3 hidden layers are necessary to compute the maximum of 5 numbers; this had only been verified with an excessive computation so far. Finally, we show that a natural generalization of the best known upper bound to maxout networks is not tight, by demonstrating that a rank-3 maxout layer followed by a rank-2 maxout layer is sufficient to represent the maximum of 7 numbers
Do ImageNet-trained models learn shortcuts?:The impact of frequency shortcuts on generalization
Get PDFFrequency shortcuts refer to specific frequency patterns that models heavily rely on for correct classification. Previous studies have shown that models trained on small image datasets often exploit such shortcuts, potentially impairing their generalization performance. However, existing methods for identifying frequency shortcuts require expensive computations and become impractical for analyzing models trained on large datasets. In this work, we propose the first approach to more efficiently analyze frequency shortcuts at a large scale. We show that both CNN and transformer models learn frequency shortcuts on ImageNet. We also expose that frequency shortcut solutions can yield good performance on out-of-distribution (OOD) test sets which largely retain texture information. However, these shortcuts, mostly aligned with texture patterns, hinder model generalization on rendition-based OOD test sets. These observations suggest that current OOD evaluations often overlook the impact of frequency shortcuts on model generalization. Future benchmarks could thus benefit from explicitly assessing and accounting for these shortcuts to build models that generalize across a broader range of OOD scenarios
Tracking Distances and Periods:Examining the Interplay between Endurance Sports and the Menstrual Cycle
No full textImprovements in sensor technology, particularly focusing on integrating sports performance metrics and health data, contributed to advancing athletic performance. Concurrently, there is an increasing focus on menstrual health within HCI that emphasises the role of technology in managing bodily experiences. Yet, despite the availability of self-tracking tools for sports and menstrual health, there is still a gap in HCI which addresses the unique needs of menstruating athletes and addresses the bidirectional effects of athletic performance and menstrual cycles. To address this gap, we conducted a diary study and interviews with N=11 runners to demonstrate how self-tracking technology embraces the interplay between athletic performance and menstrual cycles. Our results shed light on the challenges and coping strategies of menstruating athletes related to menstruation and the way tracking tools address them. This paper suggests directions for interaction designers to tackle the challenges of bidirectional impacts of athletic performance and menstrual cycles
A Frequency-Agile and Wide-Scanning CRLH Leaky-Wave Antenna for Advanced Wireless Systems
Get PDFThis letter presents a wide-angle scanning leaky-wave antenna (LWA) based on composite right-/left-handed metamaterials. The proposed LWA incorporates unit cells with a radiating slot formed by an interdigital capacitor with stepped fingers, interconnected through microstrip delay lines. The innovative radiator design achieves a wide impedance bandwidth of 4 GHz to 6.7 GHz and allows beam steering across a range of −62° to 30°. The antenna demonstrates a peak gain of 10.5 dBi and consistently high radiation efficiency, averaging 72% over the operating range. This high-performance LWA combines a simple design with a compact form factor, offering a cost-effective and easily manufacturable solution. It is specifically optimized for the C-band spectrum and is well-suited for 5G and 6G applications.</p
The link between suspect verbosity during investigative interviews and observer-rapport
Get PDFPurpose: Rapport enhances both the quantity and quality of information in investigative interviews and is recommended by multiple frameworks and training manuals. As interviewers are trained to associate rapport with more detailed responses, they are likely to assess rapport based on the amount of information provided. However, this evaluation can be skewed by the suspect’s verbosity—more elaborate accounts by the suspect may be mistaken for interview success, even when no useful information is shared. With this study, we tested whether suspect verbosity influenced observer-rapport, and judgement of perceived suspect guilt. Methods: Participants (N = 184) listened to one of three audio recordings of a mock police interview, within a between-groups design (Suspect verbosity: Low vs. Medium vs. High), and afterwards were asked to rate rapport between the suspect and the interviewer, and suspect guilt. Results: Our results show that rapport ratings statistically significantly differ between the Low and High verbosity-conditions. Participant’s also perceived suspects as being less likely to be guilty if they spoke more words, even when this speech does not provide investigation relevant information. Conclusion: If individuals base rapport and guilt judgements on verbosity rather than an objective assessment of the evidence, it may lead to sub-optimal investigative outcomes. Therefore, it is important to further investigate how interviewer training may impact our findings
Less and more than data:a Lacanian inquiry into self-formation in the age of data mining
Get PDFThis paper explores how data mining can redefine and reshape human identity, transforming the self into a fluid construct continuously shaped through ongoing profiling. To understand this transformation, we draw on Lacan’s theory of subjectivity and his conception of desire as the engine of subjectivity. Rejecting essentialist notions of the self, Lacan argues that identity is formed within—and through—social, cultural, and, as we emphasize here, technological contexts. We examine how data mining affects processes of self-formation in this technological era, using Lacanian theory as a framework to analyze its impact. We argue that data mining does not simply replicate traditional symbolic processes; rather, it introduces a different dynamic that can disrupt established modes of symbolic identification rooted in social norms, laws, and customs. This disruption may result in forms of de-identification but also opens the possibility for new types of self-identification. We propose that this transformation has a double effect: it both dissolves elements of the traditional Symbolic order and simultaneously gives rise to a new Symbolic—one that aims to define and regulate emerging identities. We believe that this tension presents both a challenge and an opportunity for contemporary processes of self-formation.</p
More Than Justifications an Analysis of Information Needs in Explanations and Motivations to Disable Personalization
Get PDFThere is consensus that algorithmic news recommenders should be explainable to inform news readers of potential risks. However, debates continue over which information users need and which stakeholders should access this information. As the debate continues, researchers also call for more control over algorithmic news recommender systems, for example, by turning off personalized recommendations. Despite this call, it is unclear the extent to which news readers will use this feature. To add nuance to the discussion, we analyzed 586 responses to two open-ended questions: i) what information needs to contribute to trustworthiness perceptions of new recommendations, and ii) whether people want the ability to turn off personalization. Our results indicate that most participants found knowing the sources of news items important for trusting a recommendation system. Additionally, more than half of the participants were inclined to disable personalization. The most common reasons to turn off personalization included concerns about bias or filter bubbles and a preference to consume generalized news. These findings suggest that news readers have different information needs for explanations when interacting with an algorithmic news recommender and that many news readers prefer to disable the usage of personalized news recommendations.</p
An analytics-based framework for military technology adoption and combat strategy
Get PDFIntroducing new technology into a military force during an ongoing conflict presents significant challenges, extending beyond logistics to include the uncertainty of how effectively soldiers can adapt to and deploy the new capabilities. This study examines how the mastery of new technology shapes the dynamics of warfare and informs effective decision-making strategies. Our methodology is grounded in a model-based approach. We begin with Lanchester's square law model, which provides a framework for analyzing modern combat scenarios involving long-range weapons. To extend this framework, we incorporate the Bass diffusion model, enabling the simultaneous examination of the progression of the conflict and the learning curve associated with the new technology. Subsequently, we utilize insights from studying the adoption of a single technology to analyze the introduction of multiple new technologies provided by different suppliers. In this context, considerations of technological effectiveness and supplier reliability become critical in making balanced procurement decisions. To support this process, we propose a Market Share Attraction model to guide decision-making effectively.</p
Venture capital dilemma:which start-up to invest in?
Get PDFPurpose: Start-ups can take an integral role in both enhancing the people’s quality of life through responding to unmet needs or enhance the level of satisfaction from the solution as well as contributing to the regional economy of their host region. However, many start-ups end up failed regarding the uncertain nature of their business, and financial failure has been mentioned as one of the principal failure factors by both entrepreneurs and scholars. Design: Having in mind that many start-ups require external financing to survive and grow, this study investigates the venture capitalists’ investment decision-making since VC funding is the most common financing method for start-ups during their scaling stage. The present study aims to develop venture capitalists’ decision-making framework for start-up selection and quantify the relative importance of each criterion. We used Multi-Criteria Decision-Making (MCDM) techniques to extract the decision criteria and the relative influence of each criterion in the final decision. Also, a case study has been presented, which validates the veracity of the derived framework. Findings: The results would help entrepreneurs to grasp a comprehensive understanding of VCs’ decision-making and enables them to optimize their businesses’ design and resource allocation. The present study would also enable both investors and investees to improve their performance. For entrepreneurs, this study provides them insight into VCs’ decision-making. The findings of this study provide valuable insights for managers looking to secure investment for their start-up. This is the first study that utilizes quantitative decision-making framework for venture capitalists. The most critical determinants of successful fundraising are identified as “prior knowledge on start-ups” followed by “team experience” and “balance team”. Originality: This research integrated “business”, “market”, “financial”, “team” and “management & operations” criteria which is unique in context of Venture Capital. Illustrating the decision criteria and the relative criteria weight, the present study enables entrepreneurs to optimize their resource allocation priorities and increase their odds of successful fundraising by focusing on the improvement in relatively more important aspects of their business. This study would also enable investors to assess their decision-making process and enhance their performance in detecting promising investment opportunities by retrospectively analyzing their failures, detecting the failure factors, and developing an objective decision-making framework.</p
Corrigendum to ‘Adverse health effects after breast cancer up to 14 years after diagnosis’ [The Breast 61 (2022) 22–28]
Get PDFThe authors regret that the printed version of the above article contained a number of errors. The correct and final version follows. The authors would like to apologise for any inconvenience caused.</p