Representativeness Research Articles

Fine resolution satellite sea surface temperatures capture the conditions experienced by corals at monthly but not daily timescales

Abstract Water temperature is a strong driver of growth, survival, and local adaptation in corals, but our knowledge of the temperatures experienced by corals on reefs worldwide remains limited. While in situ temperature loggers can provide high quality data, they are relatively expensive to place and retrieve. Alternatively, remotely sensed sea surface temperature data are globally available but may be a biased representation of the temperatures experienced by corals. Here, we compared data from 314 temperature loggers on coral reefs to the ~ 1 km2 resolution remotely sensed Multi-scale Ultra-high Resolution Sea Surface Temperature (MUR) product from NASA. We found good agreement (Pearson’s r = 0.95) between maximum monthly mean temperatures calculated from remote and in situ data, with 84% of temperatures within 0.5 °C of each other. However, remotely sensed temperature did not effectively capture sub-diel temperature fluctuations and the highest peak temperatures that may be most dangerous for corals. Predictions of in situ temperatures were significantly but weakly improved by a consideration of reef geomorphology. Ultimately, we found that remotely sensed temperatures can accurately represent the monthly conditions experienced by most corals but should be used with caution at finer temporal scales.

Conditional Generative Models for Synthetic Tabular Data: Applications for Precision Medicine and Diverse Representations.

Tabular medical datasets, like electronic health records (EHRs), biobanks, and structured clinical trial data, are rich sources of information with the potential to advance precision medicine and optimize patient care. However, real-world medical datasets have limited patient diversity and cannot simulate hypothetical outcomes, both of which are necessary for equitable and effective medical research. Fueled by recent advancements in machine learning, generative models offer a promising solution to these data limitations by generating enhanced synthetic data. This review highlights the potential of conditional generative models (CGMs) to create patient-specific synthetic data for a variety of precision medicine applications. We survey CGM approaches that tackle two medical applications: correcting for data representation biases and simulating digital health twins. We additionally explore how the surveyed methods handle modeling tabular medical data and briefly discuss evaluation criteria. Finally, we summarize the technical, medical, and ethical challenges that must be addressed before CGMs can be effectively and safely deployed in the medical field.

Pangenome graphs and their applications in biodiversity genomics.

Complete datasets of genetic variants are key to biodiversity genomic studies. Long-read sequencing technologies allow the routine assembly of highly contiguous, haplotype-resolved reference genomes. However, even when complete, reference genomes from a single individual may bias downstream analyses and fail to adequately represent genetic diversity within a population or species. Pangenome graphs assembled from aligned collections of high-quality genomes can overcome representation bias by integrating sequence information from multiple genomes from the same population, species or genus into a single reference. Here, we review the available tools and data structures to build, visualize and manipulate pangenome graphs while providing practical examples and discussing their applications in biodiversity and conservation genomics across the tree of life.

Quantifying and Learning Static Vs. Dynamic Information in Deep Spatiotemporal Networks.

There is limited understanding of the information captured by deep spatiotemporal models in their intermediate representations. For example, while evidence suggests that action recognition algorithms are heavily influenced by visual appearance in single frames, no quantitative methodology exists for evaluating such static bias in the latent representation compared to bias toward dynamics. We tackle this challenge by proposing an approach for quantifying the static and dynamic biases of any spatiotemporal model, and apply our approach to three tasks, action recognition, automatic video object segmentation (AVOS) and video instance segmentation (VIS). Our key findings are: (i) Most examined models are biased toward static information. (ii) Some datasets that are assumed to be biased toward dynamics are actually biased toward static information. (iii) Individual channels in an architecture can be biased toward static, dynamic or jointly encode a combination static and dynamic information. (iv) Most models converge to their culminating biases in the first half of training. We then explore how these biases affect performance on dynamically biased datasets. For action recognition, we propose StaticDropout, a semantically guided dropout that debiases a model from static information toward dynamics. For AVOS, we design a better combination of fusion and cross connection layers compared with previous architectures.

SEISMICITY of THE NORTH-EAST of RUSSIA in 2020

The results of seismic monitoring of the Magadan region, Chukotka Autonomous Area and the shelf of adjacent seas (Okhotsk, Chukchi, Bering and East Siberian seas) are considered. There were 12 working seismic stations in the region. The catalog of the Russian North-East earthquakes for 2020 includes information about 327 earthquakes, 114 of them are out of the region and 213 with energy classes KR=6.2–12.2 are within the region borders. As usual, most of 213 regional earthquakes (80 %) are localized in the Kolyma area. The total seismic energy released within the region’s borders was equal ΣЕ=3.4481012 J. The classification of Russian North-East earthquakes was performed using energy classes KР of T.G. Rautian’s scale. The energy representa tiveness map of earthquakes with Кmin=6–10 was built. The Magadan network of stations records without omis sions the earthquakes with Kmin=7 in Magadan oblast, with Kmin=8 in Okhotsk sea, Kmin=10 in the Chukchi. The strongest earthquake of 2020 with KR=12.2 (mb=4.3, Ms=3.7) occurred on November 1, 11h44m in the Bering Sea. Epicenters of Magadan region earthquakes were plotted on the tectonic zoning scheme. Most earthquakes are confined to the largest deep faults in the northwestern and sublatitudinal directions. All hypocenters are located within the earth's crust. In 2020, the seismicity level of the North-East of Russia according to the “SOUS09” scale was assessed as "background average" for the observation period from 1968 to 2020. Spatially, all Russian North-East earthquakes are traditionally con centrated in large seismogenic belts: Chersky, North-Okhotsk and Trans-Beringian.

Technological bias, illusory impartiality, and the injustice of hermeneutical obstruction

The algorithms and processes of modern technologies affect almost all aspects of our lives. No human being is making individual subjective choices in each, or any, instance of these processes—all cases are treated with perfect technological indifference. Hence, a commonsense assumption about the technologies we interact with, and the algorithms many of them implement, is that they are neutral and impartial. Of course, this is simply not the case. As has been extensively documented and discussed, bias exists in all facets of technology, from how and why it was first conceived and then developed to its inputs, processes and outputs. I examine instances of technological bias in visual representation and criminal justice technologies, and argue that they produce, among other epistemic concerns, a form of hermeneutical injustice by means of obstructing the deployment of the necessary and otherwise available hermeneutical resources needed to accurately understand these technologies and their effects. Such epistemic harm is inevitable, unless the relationship between our human biases and the technological processes produced by us is made explicit and significantly rethought.

Representation of a perceptual bias in the prefrontal cortex

Perception is influenced by sensory stimulation, prior knowledge, and contextual cues, which collectively contribute to the emergence of perceptual biases. However, the precise neural mechanisms underlying these biases remain poorly understood. This study aims to address this gap by analyzing neural recordings from the prefrontal cortex (PFC) of monkeys performing a vibrotactile frequency discrimination task. Our findings provide empirical evidence supporting the hypothesis that perceptual biases can be reflected in the neural activity of the PFC. We found that the state-space trajectories of PFC neuronal activity encoded a warped representation of the first frequency presented during the task. Remarkably, this distorted representation of the frequency aligned with the predictions of its Bayesian estimator. The identification of these neural correlates expands our understanding of the neural basis of perceptual biases and highlights the involvement of the PFC in shaping perceptual experiences. Similar analyses could be employed in other delayed comparison tasks and in various brain regions to explore where and how neural activity reflects perceptual biases during different stages of the trial.

Towards Context-Aware Emotion Recognition Debiasing From a Causal Demystification Perspective via De-Confounded Training.

Understanding emotions from diverse contexts has received widespread attention in computer vision communities. The core philosophy of Context-Aware Emotion Recognition (CAER) is to provide valuable semantic cues for recognizing the emotions of target persons by leveraging rich contextual information. Current approaches invariably focus on designing sophisticated structures to extract perceptually critical representations from contexts. Nevertheless, a long-neglected dilemma is that a severe context bias in existing datasets results in an unbalanced distribution of emotional states among different contexts, causing biased visual representation learning. From a causal demystification perspective, the harmful bias is identified as a confounder that misleads existing models to learn spurious correlations based on likelihood estimation, limiting the models' performance. To address the issue, we embrace causal inference to disentangle the models from the impact of such bias, and formulate the causalities among variables in the CAER task via a customized causal graph. Subsequently, we present a Contextual Causal Intervention Module (CCIM) to de-confound the confounder, which is built upon backdoor adjustment theory to facilitate seeking approximate causal effects during model training. As a plug-and-play component, CCIM can easily integrate with existing approaches and bring significant improvements. Systematic experiments on three datasets demonstrate the effectiveness of our CCIM.

Assessment and correction of representational bias in phytoliths from modern soil in the Dongting Lake Basin, China

Understanding how well topsoil phytoliths represent their aboveground plant communities, and then correcting for any representational bias, is the basis for the accurate phytolith-based reconstruction of past vegetation. We evaluated the representational bias in topsoil phytoliths at 39 sampling sites in forest, shrub, and herb communities in the Dongting Lake Basin, China. We compared the types, concentrations, and percentages of topsoil phytoliths and corresponding plant community phytoliths within the same plant community and quantified the relationships. We observed significant representational biases. Combined with the representation indices of phytoliths, we quantitatively defined the vegetation representativeness of different phytolith types in the topsoil samples. We defined four categories of representation: Associated types, Over-represented types, Under-represented types, and “Special” types. We observed differences in the vegetation representativeness of different topsoil phytolith types, and the same phytolith type also showed significant differences in vegetation representativeness between different plant communities. Additionally, using the preservation index (R-value), we corrected the representational bias of topsoil phytoliths between the different plant communities. After validation, we found that the corrected topsoil phytoliths more accurately represented quantitative changes in the aboveground plant communities. Therefore, the preservation index (R value) is an effective means of correcting the representational bias of topsoil phytoliths, and it can provide the basis for more accurate phytolith-based reconstructions of past vegetation.

The Effect of Overconfidence, Representative, Anchoring, and Availability Biases on Investment Decisions and Market Efficiency

This study looks at how behavioral biases affect investment choices and market efficiency. This research was conducted because many millennials invest on a bandwagon without having a good understanding of investment. This kind of study use the structural equation modeling analysis method. The study's findings indicate that the bias variable overconfidence behavior can strongly influence perceived market efficiency. Nevertheless, choices about investments are unaffected by the overconfidence bias, representational bias, anchoring, and availability behavior. The variable investment decision significantly influences perceived market efficiency. Investors with investment experience above five years and a high income can make a difference in investment decisions chosen by investors. This study theme's practical application relates to the findings of overconfidence bias, which has a substantial detrimental impact on investors' perceptions of market efficiency. A high degree of confidence among investors can lead to illogical judgments and disregarding all dangers, resulting in inefficient market circumstances. Keyword: anchoring, financial behavior, investment decision, market efficiency, overconfidence

Beyond risk preferences in sequential decision-making: How probability representation, sequential structure and choice perseverance bias optimal search

Sequential decision-making, where choices are made one after the other, is an important aspect of our daily lives. For example, when searching for a job, an apartment, or deciding when to buy or sell a stock, people often have to make decisions without knowing what future opportunities might arise. These situations, which are known as optimal stopping problems, involve a risk associated with the decision to either stop or continue searching. However, previous research has not consistently found a clear connection between individuals’ search behavior in these tasks and their risk preferences as measured in controlled experimental settings. In this paper, we explore how particular characteristics of optimal stopping tasks affect people’s choices, extending beyond their stable risk preferences. We find that (1) the way the underlying sampling distribution is presented (whether it is based on experience or description), (2) the sequential presentation of options, and (3) the unequal frequencies of choices to reject versus to accept significantly bias people choices. These results shed light on the complex nature of decisions that unfold sequentially and emphasize the importance of incorporating context factors when studying human decision behavior.

Assessing Methods for Adjusting Estimates of Treatment Effectiveness for Patient Nonadherence in the Context of Time-to-Event Outcomes and Health Technology Assessment: A Simulation Study.

We aim to assess the performance of methods for adjusting estimates of treatment effectiveness for patient nonadherence in the context of health technology assessment using simulation methods. We simulated trial datasets with nonadherence, prognostic characteristics, and a time-to-event outcome. The simulated scenarios were based on a trial investigating immunosuppressive treatments for improving graft survival in patients who had had a kidney transplant. The primary estimand was the difference in restricted mean survival times in all patients had there been no nonadherence. We compared generalized methods (g-methods; marginal structural model with inverse probability of censoring weighting [IPCW], structural nested failure time model [SNFTM] with g-estimation) and simple methods (intention-to-treat [ITT] analysis, per-protocol [PP] analysis) in 90 scenarios each with 1,900 simulations. The methods' performance was primarily assessed according to bias. In implementation nonadherence scenarios, the average percentage bias was 20% (ranging from 7% to 37%) for IPCW, 20% (8%-38%) for SNFTM, 20% (8%-38%) for PP, and 40% (20%-75%) for ITT. In persistence nonadherence scenarios, the average percentage bias was 26% (9%-36%) for IPCW, 26% (14%-39%) for SNFTM, 26% (14%-36%) for PP, and 47% (16%-72%) for ITT. In initiation nonadherence scenarios, the percentage bias ranged from -29% to 110% for IPCW, -34% to 108% for SNFTM, -32% to 102% for PP, and between -18% and 200% for ITT. In this study, g-methods and PP produced more accurate estimates of the treatment effect adjusted for nonadherence than the ITT analysis did. However, considerable bias remained in some scenarios. Randomized controlled trials are usually analyzed using the intention-to-treat (ITT) principle, which produces a valid estimate of effectiveness relating to the underlying trial, but when patient adherence to medications in the real world is known to differ from that observed in the trial, such estimates are likely to result in a biased representation of real-world effectiveness and cost-effectiveness.Our simulation study demonstrates that generalized methods (g-methods; IPCW, SNFTM) and per-protocol analysis provide more accurate estimates of the treatment effect than the ITT analysis does, when adjustment for nonadherence is required; however, even with these adjustment methods, considerable bias may remain in some scenarios.When real-world adherence is expected to differ from adherence observed in a trial, adjustment methods should be used to provide estimates of real-world effectiveness.

Humans underestimate the movement range of their own hands

Motor planning and motor imagery are assumed to use veridical internal representations of the biomechanical properties of our limbs. Here, we report that people underestimate their hands’ range of motion. We used two tasks probing representations of own motion range, estimation and imagery, in which participants were supposed to judge their rotational hand movement ranges. In both tasks participants’ judgments were underestimated in three out of four cardinal directions. We suggest that this representational bias provides an optimal balance between movement efficiency and safety in face of the inherently stochastic nature of movement execution.

Representation-Enhanced Status Replay Network for Multisource Remote-Sensing Image Classification.

Deep-learning-based methods are widely used in multisource remote-sensing image classification, and the improvement in their performance confirms the effectiveness of deep learning for classification tasks. However, the inherent underlying problems of deep-learning models still hinder the further improvement of classification accuracy. For example, after multiple rounds of optimization learning, representation bias and classifier bias are accumulated, which prevents the further optimization of network performance. In addition, the imbalance of fusion information among multisource images also leads to insufficient information interaction throughout the fusion process, thus making it difficult to fully utilize the complementary information of multisource data. To address these issues, a Representation-enhanced Status Replay Network (RSRNet) is proposed. First, a dual augmentation including modal augmentation and semantic augmentation is proposed to enhance the transferability and discreteness of feature representation, to reduce the impact of representation bias in the feature extractor. Then, to alleviate the classifier bias and maintain the stability of the decision boundary, a status replay strategy (SRS) is built to regulate the learning and optimization of the classifier. Finally, aiming to improve the interactivity of modal fusion, a novel cross-modal interactive fusion (CMIF) method is employed to jointly optimize the parameters of different branches by combining multisource information. Quantitative and qualitative results on three datasets demonstrate the superiority of RSRNet in multisource remote-sensing image classification, and its outperformance compared with other state-of-the-art methods.

Evaluating the impact of uncertainty in ground motion forecasts for post-earthquake impact modeling applications

The US Geological Survey’s (USGS) ShakeMap system provides a rapid characterization of strong ground shaking in areas directly affected by an earthquake. This study focuses on studying the aggregate effects of macroseismic shaking estimates from ShakeMap, expressed in terms of modified Mercalli intensity (MMI), when accounting for the uncertainty in forecasted ground motions. We use a Monte Carlo approach to generate numerous spatially correlated realizations of ground motions by utilizing a combination of circulant embedding and kriging techniques for efficiently handling the correlations. We then assessed the aggregate effects of shaking by looking at bin counts across these realizations. We demonstrate that the aggregate shaking regarding the mean macroseismic intensity estimates (from the ShakeMap output) is a biased representation of the aggregate shaking when shaking uncertainty is included. Incorporating shaking uncertainty can help to improve various downstream earthquake impact applications, such as the USGS Prompt Assessment of Global Earthquakes for Response (PAGER) overall earthquake fatality distribution or estimates of shaking-induced ground failure impacts from consequential earthquakes.

Supplemental Material for Regional Gender Bias and Year Predict Gender Representation on Civil Trial Teams

Supplemental Material for Regional Gender Bias and Year Predict Gender Representation on Civil Trial Teams

Effect of attention on ensemble perception: Comparison between exogenous attention, endogenous attention, and depth.

Ensemble perception is an important ability of human beings that allows one to extract summary information for scenes and environments that contain information that far exceeds the processing limit of the visual system. Although attention has been shown to bias ensemble perception, two important questions remain unclear: (1) whether direct manipulations on different types of spatial attention could produce similar effects on ensembles and (2) whether factors potentially influencing the attention distribution, such as depth perception, could evoke an indirect effect of attention on ensemble representation. This study aims to address these questions. In Experiments 1 and 2, two types of precues were used to evoke exogenous and endogenous attention, respectively, and the ensemble color perceptions were examined. We found that both exogenous and endogenous attention biased ensemble representation towards the attended items, and the latter produced a greater effect. In Experiments 3 and 4, we examined whether depth perception could affect color ensembles by indirectly influencing attention allocation in 3D space. The items were separated in two depth planes, and no explicit cues were applied. The results showed that color ensemble was biased to closer items when depth information was task relevant. This suggests that ensemble perception is naturally biased in 3D space, probably through the mechanism of attention. Computational modeling consistently showed that attention exerted a direct shift on the ensemble statistics rather than averaging the feature values over the cued and noncued items, providing evidence against an averaging process of individual perception.

Less can be more: representational vs. stereotypical gender bias in facial expression recognition

AbstractMachine learning models can inherit biases from their training data, leading to discriminatory or inaccurate predictions. This is particularly concerning with the increasing use of large, unsupervised datasets for training foundational models. Traditionally, demographic biases within these datasets have not been well-understood, limiting our ability to understand how they propagate to the models themselves. To address this issue, this paper investigates the propagation of demographic biases from datasets into machine learning models. We focus on the gender demographic component, analyzing two types of bias: representational and stereotypical. For our analysis, we consider the domain of facial expression recognition (FER), a field known to exhibit biases in most popular datasets. We use Affectnet, one of the largest FER datasets, as our baseline for carefully designing and generating subsets that incorporate varying strengths of both representational and stereotypical bias. Subsequently, we train several models on these biased subsets, evaluating their performance on a common test set to assess the propagation of bias into the models’ predictions. Our results show that representational bias has a weaker impact than expected. Models exhibit a good generalization ability even in the absence of one gender in the training dataset. Conversely, stereotypical bias has a significantly stronger impact, primarily concentrated on the biased class, although it can also influence predictions for unbiased classes. These results highlight the need for a bias analysis that differentiates between types of bias, which is crucial for the development of effective bias mitigation strategies.

Bias Mitigation and Representation Optimization for Noise-Robust Cross-modal Retrieval

The remarkable progress in cross-modal retrieval relies on accurately-annotated multimedia datasets. In practice, most existing datasets used for training cross-modal retrieval models are automatically collected from the Internet to reduce data collection costs. However, it inevitably contains mismatched pairs, i.e. , noisy correspondences, thus degrading the model performance. Recent advances utilize the predicted similarity distribution of individual samples for noise validation and correction, which easily faces two challenging dilemmas: 1) confirmation bias and 2) unstable performance with increasing noise. In light of the above, we propose a generalized B ias M itigation and R epresentation O ptimization framework (BMRO). Specifically, we propose a Bias Estimator (BE) to estimate the unbiased confidence factor of a sample by contrasting it against its nearest neighbors. Unbiased confidence factor can precisely adjust sample contribution and enhance accurate sample division. This facilitates the Adaptive Representation Optimizer (ARO) in providing tailored optimization strategies for clean and noisy samples. ARO performs contrastive learning between clean samples and generated hard samples, thus promoting the generalizability and robustness of the representation. Besides, it utilizes complementary learning to reduce incorrect guidance from noisy samples. Extensive experiments on five visual-text benchmarks verify that our BMRO can significantly improve the matching accuracy and performance stability against noisy correspondences.

Beyond the stereotypes: Artificial Intelligence image generation and diversity in anesthesiology.

Artificial Intelligence (AI) is increasingly being integrated into anesthesiology to enhance patient safety, improve efficiency, and streamline various aspects of practice. This study aims to evaluate whether AI-generated images accurately depict the demographic racial and ethnic diversity observed in the Anesthesia workforce and to identify inherent social biases in these images. This cross-sectional analysis was conducted from January to February 2024. Demographic data were collected from the American Society of Anesthesiologists (ASA) and the European Society of Anesthesiology and Intensive Care (ESAIC). Two AI text-to-image models, ChatGPT DALL-E 2 and Midjourney, generated images of anesthesiologists across various subspecialties. Three independent reviewers assessed and categorized each image based on sex, race/ethnicity, age, and emotional traits. A total of 1,200 images were analyzed. We found significant discrepancies between AI-generated images and actual demographic data. The models predominantly portrayed anesthesiologists as White, with ChatGPT DALL-E2 at 64.2% and Midjourney at 83.0%. Moreover, male gender was highly associated with White ethnicity by ChatGPT DALL-E2 (79.1%) and with non-White ethnicity by Midjourney (87%). Age distribution also varied significantly, with younger anesthesiologists underrepresented. The analysis also revealed predominant traits such as "masculine, ""attractive, "and "trustworthy" across various subspecialties. AI models exhibited notable biases in gender, race/ethnicity, and age representation, failing to reflect the actual diversity within the anesthesiologist workforce. These biases highlight the need for more diverse training datasets and strategies to mitigate bias in AI-generated images to ensure accurate and inclusive representations in the medical field.