Search

Abstract This paper addresses the often overlooked issue of fairness in the autonomous driving domain, particularly in vision-based perception and prediction systems, which play a pivotal role in the overall functioning of Autonomous Vehicles (AVs). We focus our analysis on biases present in some of the most commonly used visual datasets for training person and vehicle detection systems. We introduce an annotation methodology and a specialised annotation tool, both designed to annotate protected attributes of agents in visual datasets. We validate our methodology through an inter-rater agreement analysis and provide the distribution of attributes across all datasets. These include annotations for the attributes age, sex, skin tone, group, and means of transport for more than 90K people, as well as vehicle type, colour, and car type for over 50K vehicles. Generally, diversity is very low for most attributes, with some groups, such as children, wheelchair users, or personal mobility vehicle users, being extremely underrepresented in the analysed datasets. The study contributes significantly to efforts to consider fairness in the evaluation of perception and prediction systems for AVs. This paper follows reproducibility principles. The annotation tool, scripts and the annotated attributes can be accessed publicly at https://github.com/ec-jrc/humaint_annotator .

This paper explores the relationship between media, government, and terrorism. It investigates what terrorists want from the media, what governments seek from the media when covering terrorism, and what the media aims to achieve in its coverage of terrorism. The paper discusses coverage versus boycott, extortion and exploitation, ethical considerations, the war on terrorism and public diplomacy, current trends in terrorist operations, and attacks on journalists and media institutions, the Media in the West, and the Middle East dealings with terrorism. The purpose of the study is to investigate the complex relationship of three parties (terrorists, media, and government) with different objectives and agendas. The problem of media and terrorism lies in the conflicting interests and goals of these three different entities. The author used several case studies related to the problem of this study to highlight the conflictual approach to terrorism and how each entity looks at it, making the equation of terrorism, government, and media complex, challenging, and extremely sensitive that must be managed with caution. Both media and political institutions must consider several factors such as national security, public interest, economic, political, and ethical considerations. The relationship between terrorism, media and governments are characterized by a set of challenges, such as the public right to know and the potential for extortion and exploitation by terrorist groups, and the attempts from the governments to set the agendas. How can the media serve the truth without falling into the trap of extortion, exploitation, without succumbing to the pursuit of journalistic scoops, and sensationalism and without falling into the pressure of control and pressures from the governments?

The International Committee of Medical Journal Editors (ICMJE) guidelines are widely employed as an set of ethical standards for biomedical journals, and thus for biomedical researchers. In this paper, we revisit the topic of acknowledgements in academic papers, noting that the former serve as a lesser form of recognition relative to authorship. We note the possible existence of bias, such as a power imbalance due to a status imbalance, as well as the risk of “ghost” acknowledgements. To further ground our ideas, we turned to ChatGPT-4 for input, noting some curious and informative supplementary findings. Curiously, ChatGPT-4 offered a set of recommendations and guidance, comparable to those of the ICMJE.

In this paper we investigate a model of consensus decision making [Hartnett A. T., et al., Phys. Rev. Lett., 2016, 116, 038701] following a statistical physics approach presented in [Sarkanych P., et al., Phys. Biol., 2023, 20, 045005]. Within this approach, the temperature serves as a measure of fluctuations, not considered before in the original model. Here, we discuss the model on a complete graph. The main goal of this paper is to show that an analytical description may lead to a very rich phase behaviour, which is usually not expected for a complete graph. However, the variety of individual agent (spin) features - their inhomogeneity and bias strength - taken into account by the model leads to rather non-trivial collective effects. We show that the latter may emerge in a form of continuous or abrupt phase transitions sometimes accompanied by re-entrant and order-parameter flipping behaviour. In turn, this may lead to appealing interpretations in terms of social decision making. We support analytical predictions by numerical simulation. Moreover, while analytical calculations are performed within an equilibrium statistical physics formalism, the numerical simulations add yet another dynamical feature - local non-linearity or conformity of the individual to the opinion of its surroundings. This feature appears to have a strong impact both on the way in which an equilibrium state is approached as well as on its characteristics.

Abstract Background Medical text, as part of an electronic health record, is an essential information source in healthcare. Although natural language processing (NLP) techniques for medical text are developing fast, successful transfer into clinical practice has been rare. Especially the hospital domain offers great potential while facing several challenges including many documents per patient, multiple departments and complex interrelated processes. Methods In this work, we survey relevant literature to identify and classify approaches which exploit NLP in the clinical context. Our contribution involves a systematic mapping of related research onto a prototypical patient journey in the hospital, along which medical documents are created, processed and consumed by hospital staff and patients themselves. Specifically, we reviewed which dataset types, dataset languages, model architectures and tasks are researched in current clinical NLP research. Additionally, we extract and analyze major obstacles during development and implementation. We discuss options to address them and argue for a focus on bias mitigation and model explainability. Results While a patient’s hospital journey produces a significant amount of structured and unstructured documents, certain steps and documents receive more research attention than others. Diagnosis, Admission and Discharge are clinical patient steps that are researched often across the surveyed paper. In contrast, our findings reveal significant under-researched areas such as Treatment, Billing, After Care, and Smart Home. Leveraging NLP in these stages can greatly enhance clinical decision-making and patient outcomes. Additionally, clinical NLP models are mostly based on radiology reports, discharge letters and admission notes, even though we have shown that many other documents are produced throughout the patient journey. There is a significant opportunity in analyzing a wider range of medical documents produced throughout the patient journey to improve the applicability and impact of NLP in healthcare. Conclusions Our findings suggest that there is a significant opportunity to leverage NLP approaches to advance clinical decision-making systems, as there remains a considerable understudied potential for the analysis of patient journey data.

This paper explores recent advancements and implications of artificial intelligence (AI) technology, with a specific focus on Large Language Models (LLMs) like ChatGPT 3.5, within the realm of higher education. Through a review of the academic literature, this paper highlights the unprecedented growth of these models and their wide-reaching impact across various sectors. The discussion sheds light on the complex issues and potential benefits presented by LLMs, providing a overview of the field's current state. In the context of higher education, the paper explores the challenges and opportunities posed by LLMs. These include issues related to educational assessment, potential threats to academic integrity, privacy concerns, the propagation of misinformation, EDI aspects, copyright concerns and inherent biases within the models. While these challenges are multifaceted and significant, the paper emphasizes the availability of strategies to address them effectively and facilitate the successful adoption of LLMs in educational settings. Furthermore, the paper recognises the potential opportunities to transform higher education. It emphasises the need to update assessment policies, develop guidelines for staff and students, scaffold AI skills development, and find ways to leverage technology in the classroom. By proactively pursuing these steps, higher education institutions (HEIs) can harness the full potential of LLMs while managing their adoption responsibly. In conclusion, the paper urges HEIs to allocate resources to handle the adoption of LLMs effectively. This includes ensuring staff AI readiness and taking steps to modify their study programmes to align with the evolving educational landscape influenced by emerging technologies.

This paper introduces a new method to estimate the population variance of a study variable in stratified successive sampling over two occasions, while accounting for random non-response. The method uses a logarithmic type estimator and leverages information from a highly positively correlated auxiliary variable. The paper also presents calibrated weights for the new estimator and examines its properties through numerical and simulation studies. The results indicate that the suggested estimator is more effective than the standard estimator for estimating the population variance.

The COVID-19 pandemic has significantly impacted global health, resulting in numerous fatalities and presenting substantial challenges to national healthcare systems due to a sharp increase in cases. Key to managing this crisis is the rapid and accurate identification of COVID-19 infections, a task that can be enhanced with Machine Learning (ML) techniques. However, ML applications can also generate biased and potentially unfair outcomes for certain demographic groups. This paper introduces a ML model designed for detecting both COVID-19 cases and biases associated with specific patient attributes. The model employs Decision Tree and XGBoost algorithms for case detection, while bias analysis is performed using the DALEX library, which focuses on protected attributes such as age, gender, race, and ethnicity. DALEX works by creating an "explainer" object that represents the model, enabling exploration of the model's functions without requiring in-depth knowledge of its workings. This approach helps pinpoint influential attributes and uncover potential biases within the model. Model performance is assessed through accuracy metrics, with the Decision Tree algorithm achieving the highest accuracy at 99% following Bayesian hyperparameter optimization. However, high accuracy does not ensure fairness, as biases related to protected attributes may still persist.

The impact of non-participation in surveys depends on how many people don't respond and the extent to which those who don't differ from those who do on key questions. In health research, non-participation can skew results, affecting our understanding of disease and healthcare needs. Understanding why people don't respond to surveys is crucial for accurately gauging health issues. Despite the popularity of surveys in health research, the problem of non-response bias persists due to low response rates. This paper explores various methods, their strengths, and weaknesses, to address non-response issues in health surveys.

Background We explored associations between possible demographic and socioeconomic causes of religious/spiritual beliefs and behaviours (RSBB) in the offspring generation of the Avon Longitudinal Study of Parents and Children (ALSPAC). Methods We examined approximately 4,450 offspring aged 28 years with RSBB data from a prospective birth cohort study (ALSPAC) in Southwest England. Three RSBB outcome measures were assessed: religious belief (belief in God/a divine power; yes/not sure/no), religious affiliation (Christian/none/other) and religious attendance (frequency of attendance at a place of worship). We explored age- and sex-adjusted associations between 35 demographic and socioeconomic exposures and each of the three RSBB outcomes using multinomial regression. Exposure-sex interactions were also examined. Results Some sociodemographic factors were associated with RSBB in this cohort; for instance, being female and from an ethnicity other than White were associated with increased religiosity across all domains. For many other exposures, however, associations were frequently null or inconsistent, often depending on the specific exposure and outcome combination. As an example, higher educational attainment was associated with higher rates of religious attendance, but not with religious belief or affiliation; in contrast, higher income was associated with lower levels of religiosity. No consistent interactions between sex and the exposures on RSBB were found. Effect sizes were also rather weak, with most pseudo-R 2 values below 0.5% and a maximum of 1.2%. Conclusions The results highlight that several demographic and socioeconomic factors are associated with RSBB in this cohort. However, the number of these associations, and their magnitude, is smaller than comparable results from the parental generation of these offspring, suggesting that patterns of sociodemographic factors associated with RSBB differ between these generations. In addition to describing these associations, this paper will help inform future studies using these data, particularly regarding the choice of potential sociodemographic confounders.

The article discusses one of the most widely used distribution laws for reliability analysis – Weibull distribution. It describes a wide range of processes for all stages of the life cycle of technical objects, including yield stress of steel distribution and failures in the reliability theory regarding the wide range of technical objects (e.g., metal cutting tools, bearings, compressors, and wheels). A significant number of works are devoted to evaluating distribution law parameters based on empirical data in search of the most precise one, ignoring the probabilistic character of the parameters themselves. Parameters may have a relatively wide confidence range, which can be considered the parameter estimation error compared to biases of parameters estimated by different methods. Moreover, many approaches should be used for certain selection volumes, including comprehensive calculating procedures. Instead, this paper suggested and statistically confirmed a universal simplified approach. It demands a minimal set of data and connects the shape and scale parameters of the Weibull distribution with the variation coefficient as one of the leading statistical characteristics. This approach does not demand variational sequence arrangement. Nevertheless, it is supposed to be quite efficient for the engineering practice of reliability analysis. The adequacy of the results was confirmed using generated selections analysis and experimental data on cutting tool reliability. Within the achieved results, it was also demonstrated that the variation coefficient reflects not only selection stability and variable volatility degree, which are its main aim, but the cause of failure as well.

Artificial Intelligence (AI) systems can introduce biases that lead to unreliable outcomes and, in the worst-case scenarios, perpetuate systemic and discriminatory results when deployed in the real world. While significant efforts have been made to create bias detection methods, developing reliable and comprehensive documentation artifacts also makes for valuable resources that address bias and aid in minimizing the harms associated with AI systems. Based on compositional design patterns, this paper introduces a documentation approach using a hybrid AI system to prompt the identification and traceability of bias in datasets and predictive AI models. To demonstrate the effectiveness of our approach, we instantiate our pattern in two implementations of a hybrid AI system. One follows an integrated approach and performs fine-grained tracing and documentation of the AI model. In contrast, the other hybrid system follows a principled approach and enables the documentation and comparison of bias in the input data and the predictions generated by the model. Through a use-case based on Fake News detection and an empirical evaluation, we show how biases detected during data ingestion steps (e.g., label, over-representation, activity bias) affect the training and predictions of the classification models. Concretely, we report a stark skewness in the distribution of input variables towards the Fake News label, we uncover how a predictive variable leads to more constraints in the learning process, and highlight open challenges of training models with unbalanced datasets. A video summarizing this work is available online (https://youtu.be/v2GfIQPAy_4?si=BXtWOf97cLiZavyu),and the implementation is publicly available on GitHub (https://github.com/SDM-TIB/DocBiasKG).

Mixup is a powerful data augmentation strategy that has been shown to improve the generalization and adversarial robustness of machine learning classifiers, particularly in computer vision applications. Despite its simplicity and effectiveness, the impact of Mixup on the fairness of a model has not been thoroughly investigated yet. In this paper, we demonstrate that Mixup can perpetuate or even exacerbate bias presented in the training set. We provide insight to understand the reasons behind this behavior and propose GBMix, a group-balanced Mixup strategy to train fair classifiers. It groups the dataset based on their attributes and balances the Mixup ratio between the groups. Through the reorganization and balance of Mixup among groups, GBMix effectively enhances both average and worst-case accuracy concurrently. We empirically show that GBMix effectively mitigates bias in the training set and reduces the performance gap between groups. This effect is observed across a range of datasets and networks, and GBMix outperforms all the state-of-the-art methods.

Although cross-cultural interactions and cultural bubbles have been researched extensively in tourism, these issues have often been viewed one-sidedly. More precisely, in contemporary literature, cultural bubbles are mostly considered one-sided biases that prevent tourists from fully enjoying the culture of their hosts. Besides that, it is important to bear in mind that a tourist offer itself is also formed from the perspectives of tourism experience designers and tourism managers. Therefore, it is important to consider how the perception of a tourist offer is distorted by looking from inside the bubble of tourist culture. This paper highlights managers’ cultural biases in designing tourist experiences, which have often been overlooked in cross-cultural tourism studies. It is crucial to have this discussion now so that the global tourist industry can continue to provide local experiences, as tourism managers and employees in the sector are increasingly interacting with people from diverse cultural contexts. This might be beneficial for the cultural expression of destinations, on the one hand, as well as for providing more valuable experiences for tourists, on the other.

The immersion of virtual reality (VR) can impact user perceptions in numerous forms, even racial bias and embodied experiences. These effects are often limited to head-mounted displays (HMDs) and other immersive technologies that may not be inclusive to the general population. This paper investigates racial bias and embodiment on a less immersive but more accessible medium: desktop VR. A population of participants (n = 158) participated in a desktop simulation where they embodied a virtual avatar and interacted with virtual humans to determine if desktop embodiment is induced and if there is a resulting effect on racial bias. Our results indicate that desktop embodiment can be induced at low levels, as measured by an embodiment questionnaire. Furthermore, one’s implicit bias may actually influence embodiment, and the experience and perceptions of a desktop VR simulation can be improved through embodied avatars. We discuss these findings and their implications in the context of stereotype activation and existing literature in embodiment.

The gap between genomics and phenomics is narrowing. The rate at which it is narrowing, however, is being slowed by improper statistical comparison of methods. Quantification using Pearson’s correlation coefficient (r) is commonly used to assess method quality, but it is an often misleading statistic for this purpose as it is unable to provide information about the relative quality of two methods. Using r can both erroneously discount methods that are inherently more precise and validate methods that are less accurate. These errors occur because of logical flaws inherent in the use of r when comparing methods, not as a problem of limited sample size or the unavoidable possibility of a type I error. A popular alternative to using r is to measure the limits of agreement (LOA). However both r and LOA fail to identify which instrument is more or less variable than the other and can lead to incorrect conclusions about method quality. An alternative approach, comparing variances of methods, requires repeated measurements of the same subject, but avoids incorrect conclusions. Variance comparison is arguably the most important component of method validation and, thus, when repeated measurements are possible, variance comparison provides considerable value to these studies. Statistical tests to compare variances presented here are well established, easy to interpret and ubiquitously available. The widespread use of r has potentially led to numerous incorrect conclusions about method quality, hampering development, and the approach described here would be useful to advance high throughput phenotyping methods but can also extend into any branch of science. The adoption of the statistical techniques outlined in this paper will help speed the adoption of new high throughput phenotyping techniques by indicating when one should reject a new method, outright replace an old method or conditionally use a new method.

Remotely sensed data are frequently used for predicting and mapping ecosystem characteristics, and spatially explicit wall-to-wall information is sometimes proposed as the best possible source of information for decision-making. However, wall-to-wall information typically relies on model-based prediction, and several features of model-based prediction should be understood before extensively relying on this type of information. One such feature is that model-based predictors can be considered both unbiased and biased at the same time, which has important implications in several areas of application. In this discussion paper, we first describe the conventional model-unbiasedness paradigm that underpins most prediction techniques using remotely sensed (or other) auxiliary data. From this point of view, model-based predictors are typically unbiased. Secondly, we show that for specific domains, identified based on their true values, the same model-based predictors can be considered biased, and sometimes severely so.We suggest distinguishing between conventional model-bias, defined in the statistical literature as the difference between the expected value of a predictor and the expected value of the quantity being predicted, and design-bias of model-based estimators, defined as the difference between the expected value of a model-based estimator and the true value of the quantity being predicted. We show that model-based estimators (or predictors) are typically design-biased, and that there is a trend in the design-bias from overestimating small true values to underestimating large true values. Further, we give examples of applications where this is important to acknowledge and to potentially make adjustments to correct for the design-bias trend. We argue that relying entirely on conventional model-unbiasedness may lead to mistakes in several areas of application that use predictions from remotely sensed data.