Model Bias Research Articles

Convection-permitting climate model representation of severe convective wind gusts and future changes in southeastern Australia

Abstract. Previous research has suggested that the frequency and intensity of surface hazards associated with thunderstorms and convection, such as severe convective winds (SCWs), could potentially change in a future climate due to global warming. However, because of the small spatial scales associated with SCWs, they are unresolved in global climate models, and future climate projections are uncertain. Here, we evaluate the representation of SCW events in a convection-permitting climate model (Bureau of Meteorology Atmospheric Regional Projections for Australia, BARPAC-M) run over southeastern Australia for the months of December–February. We also assess changes in SCW event frequency in a projected future climate for the year 2050 and compare this with an approach based on identifying large-scale environments favourable for SCWs from a regional parent model (BARPA-R). This is done for three different types of SCW events that have been identified in this region, based on clustering of the large-scale environment. Results show that BARPAC-M representation of the extreme daily maximum wind gust distribution is improved relative to the gust distribution simulated by the regional parent model. This is due to the high spatial resolution of BARPAC-M output, as well as partly resolving strong and short-lived gusts associated with convection. However, BARPAC-M significantly overestimates the frequency of simulated SCW events, particularly in environments having steep low-level temperature lapse rates. A future decrease in SCW frequency under conditions with steep lapse rates is projected by BARPAC-M, along with less frequent favourable large-scale environments. In contrast, an increase in SCW frequency is projected under conditions of high surface moisture, with more frequent favourable large-scale environments. Therefore, overall changes in SCWs for this region remain uncertain, due to different responses between event types, combined with historical model biases.

A toolbox for surfacing health equity harms and biases in large language models.

Large language models (LLMs) hold promise to serve complex health information needs but also have the potential to introduce harm and exacerbate health disparities. Reliably evaluating equity-related model failures is a critical step toward developing systems that promote health equity. We present resources and methodologies for surfacing biases with potential to precipitate equity-related harms in long-form, LLM-generated answers to medical questions and conduct a large-scale empirical case study with the Med-PaLM 2 LLM. Our contributions include a multifactorial framework for human assessment of LLM-generated answers for biases and EquityMedQA, a collection of seven datasets enriched for adversarial queries. Both our human assessment framework and our dataset design process are grounded in an iterative participatory approach and review of Med-PaLM 2 answers. Through our empirical study, we find that our approach surfaces biases that may be missed by narrower evaluation approaches. Our experience underscores the importance of using diverse assessment methodologies and involving raters of varying backgrounds and expertise. While our approach is not sufficient to holistically assess whether the deployment of an artificial intelligence (AI) system promotes equitable health outcomes, we hope that it can be leveraged and built upon toward a shared goal of LLMs that promote accessible and equitable healthcare.

Assessing Climate Extremes in Dynamical Downscaling Simulations Driven by a Novel Bias‐Corrected CMIP6 Data

AbstractDynamical downscaling is a widely‐used approach for generating regional projections of climate extremes at a finer scale. However, the systematic bias of the global climate model (GCM) generally degrades the reliability of projections. Recently, novel bias‐corrected CMIP6 data was generated using a mean‐variance‐trend (MVT) bias correction method for dynamical downscaling simulation. To validate the effectiveness of this data in the dynamical downscaling simulation of climate extremes, we carry out three Weather Research and Forecasting (WRF) simulations over Asia‐western North Pacific with a 25 km grid spacing from 1980 to 2014. The dynamical downscaling simulations driven by the raw GCM data set (hereafter WRF_GCM) and the bias‐corrected GCM (hereafter WRF_GCMbc) were assessed against the simulation driven by the European Center for Medium‐Range Weather Forecasts Reanalysis 5 data set. The results indicate that the MVT bias correction significantly improves the climatological mean and inter‐annual variability of downscaled climate extreme indices. In terms of the climatological mean, the WRF_GCMbc shows a 25%–82% decrease in root mean square errors (RMSEs) against the WRF_GCM. As for the inter‐annual variability, the MVT bias correction can improve the downscaling simulation of almost all precipitation extreme indices and 70% of the temperature extreme indices, characterized by the RMSEs' reduction of 1%–58%. The improvements of the climate extremes in terms of the climatological mean in the WRF_GCMbc primarily stem from the improved large‐scale circulation and ocean evaporation in the WRF, which in turn improves the downscaled precipitation and 2 m temperature through the advection, radiation, and surface energy exchange process.

Introduction of a Framework for the Integration of a Kinematic Robot Arm Model in an Artificial Neural Network - Extended Kalman Filter Approach

The aim of this paper is to introduce a model in which systematic effects can be assigned according to their origin or mode of action. The approach intends to improve the positional accuracy of a robot arm. We show the impact of unaccounted model biases on estimated parameters when applying sequential approaches and conclude the necessity of jointly determining all influencing variables. Therefore, we propose a simultaneous estimation of transformation parameters, robot’s kinematic parameters and non-geometric parameters modelled by an artificial neural network (ANN) in further consequence. Thus, the main contribution of this paper is a new approach of the simultaneous estimation of the geometric and non-geometric components of a robot arm model. The integration of the geometric model (transformations, kinematic robot model) with the non-geometric one (ANN) is realised in the extended Kalman filter. The functionality of the algorithm has been proven on simulated data. The adaptive behaviour of machine learning approaches is made possible by an additional iteration of the ANN. The initialisation of the ANN parameters must not deviate from the nominal parameters by more than 10% so that the ANN can learn the non-geometric part. In this setup, the robot arm position corrections are reduced by 32.5%. A final sensitivity analysis proves the estimability of most kinematic parameters in the course of a future adaptive extension of the approach.

High‐dimensional ensemble Kalman filter with localization, inflation, and iterative updates

AbstractAccurate estimation of forecast‐error covariance matrices is an essential step in data assimilation, which becomes a challenging task for high‐dimensional data assimilation. The standard ensemble Kalman filter (EnKF) may diverge due to both the limited ensemble size and the model bias. In this article, we propose to replace the sample covariance in the EnKF with a statistically consistent high‐dimensional tapering covariance matrix estimator to counter the estimation problem under high dimensions. A high‐dimensional EnKF scheme combining covariance localization with the inflation method and an iterative update structure is developed. The proposed assimilation scheme is tested on the Lorenz‐96 model with spatially correlated observation systems. The results demonstrate that the proposed method could improve the assimilation performance under multiple settings.

Artificial intelligence in psychiatric diagnosis: challenges and opportunities in the era of machine learning

The integration of artificial intelligence (AI) into psychiatric diagnosis heralds a new era in mental health care, offering unprecedented opportunities to enhance diagnostic accuracy, personalize treatment, and streamline clinical workflows. A systematic approach was utilized, adhering to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. This literature review explores the current state of AI in psychiatric diagnosis, highlighting key technologies such as machine learning, natural language processing, and deep learning. We discuss the application of these technologies across various psychiatric disorders, including depression, anxiety, and schizophrenia. While AI holds immense promise, significant challenges remain, including issues of data privacy, model bias, and the clinical validation of AI tools. Furthermore, ethical and regulatory considerations must be addressed to ensure responsible implementation. This review also examines the potential future directions of AI in psychiatry, emphasizing the importance of collaboration between AI systems and human clinicians. As the field evolves, AI has the potential to transform psychiatric practice, offering new avenues for early detection, personalized care, and therapeutic monitoring.

Fairness in AI-Driven Oncology: Investigating Racial and Gender Biases in Large Language Models.

Large language model (LLM) chatbots have many applications in medical settings. However, these tools can potentially perpetuate racial and gender biases through their responses, worsening disparities in healthcare. With the ongoing discussion of LLM chatbots in oncology and the widespread goal of addressing cancer disparities, this study focuses on biases propagated by LLM chatbots in oncology. Chat Generative Pre-trained Transformer (Chat GPT; OpenAI, San Francisco, CA, USA) was asked to determine what occupation a generic description of "assesses cancer patients" would correspond to for different demographics. Chat GPT, Gemini (Alphabet Inc., Mountain View, CA, USA), and Bing Chat (Microsoft Corp., Redmond, WA, USA) were prompted to provide oncologist recommendations in the top U.S. cities and demographic information (race, gender) of recommendations was compared against national distributions. Chat GPT was also asked to generate a job description for oncologists with different demographic backgrounds. Finally, Chat GPT, Gemini, and Bing Chat were asked to generate hypothetical cancer patients with race, smoking, and drinking histories. LLM chatbots are about two times more likely to predict Blacksand Native Americans as oncology nurses than oncologists, compared to Asians (p < 0.01 and < 0.001, respectively). Similarly, they are also significantly more likely to predict females than males as oncology nurses (p < 0.001). Chat GPT's real-world oncologist recommendations overrepresent Asians by almost double and underrepresent Blacks by double and Hispanics by seven times. Chatbots also generate different job descriptions based on demographics, including cultural competency and advocacy and excluding treatment administration for underrepresented backgrounds. AI-generated cancer cases are not fully representative of real-world demographic distributions and encode stereotypes on substance abuse, such as Hispanics having a greater proportion of smokers than Whites by about 20% in Chat GPT breast cancer cases. To our knowledge, this is the first study of its kind to investigate racial and gender biases of such a diverse set of AI chatbots, and that too, within oncology. The methodology presented in this study provides a framework for targeted bias evaluation of LLMs in various fields across medicine.

Motion adaptation induced object position bias in macaque IT and SlowFast video recognition models

Motion adaptation induced object position bias in macaque IT and SlowFast video recognition models

Semi-Supervised Remote Sensing Building Change Detection with Joint Perturbation and Feature Complementation

The timely updating of the spatial distribution of buildings is essential to understanding a city’s development. Deep learning methods have remarkable benefits in quickly and accurately recognizing these changes. Current semi-supervised change detection (SSCD) methods have effectively reduced the reliance on labeled data. However, these methods primarily focus on utilizing unlabeled data through various training strategies, neglecting the impact of pseudo-changes and learning bias in models. When dealing with limited labeled data, abundant low-quality pseudo-labels generated by poorly performing models can hinder effective performance improvement, leading to the incomplete recognition results of changes to buildings. To address this issue, we propose a feature multi-scale information interaction and complementation semi-supervised method based on consistency regularization (MSFG-SemiCD), which includes a multi-scale feature fusion-guided change detection network (MSFGNet) and a semi-supervised update method. Among them, the network facilitates the generation of multi-scale change features, integrates features, and captures multi-scale change targets through the temporal difference guidance module, the full-scale feature fusion module, and the depth feature guidance fusion module. Moreover, this enables the fusion and complementation of information between features, resulting in more complete change features. The semi-supervised update method employs a weak-to-strong consistency framework to achieve model parameter updates while maintaining perturbation invariance of unlabeled data at both input and encoder output features. Experimental results on the WHU-CD and LEVIR-CD datasets confirm the efficacy of the proposed method. There is a notable improvement in performance at both the 1% and 5% levels. The IOU in the WHU-CD dataset increased by 5.72% and 6.84%, respectively, while in the LEVIR-CD dataset, it improved by 18.44% and 5.52%, respectively.

Biased Mediators in Conflict Resolution

Abstract One of the most important and disputed questions within the fields of international relations and conflict studies concerns the issue of mediator impartiality. Should mediators be biased—supportive of one but not both of the main disputants—or should mediators always be impartial? This paper contributes to this debate by studying the effectiveness of mediation with regard to the role of mediator bias in a game-theoretic model of cheap talk. This study shows that the institutional design of a mediation process is affected by two factors: the relative degree of conflict and the incentives to misrepresent private information. We find that a necessary (but not always sufficient) condition for the success of a mediation process is a sufficiently low likelihood of a misrepresentation problem. If in addition, the relative degree of conflict is low enough, mediation is effective, and the institutional design of a mediation process is the same regardless of the bias. Otherwise, the design will be quite different depending on the direction of the bias. (JEL D63, D74, D82)

A spectral bias-error stepwise correction method of plasma image-spectrum fusion based on deep learning for improving the performance of LIBS

Poor spectral stability seriously hinders the wide application of laser-induced breakdown spectroscopy (LIBS), so how to improve its stability is the focus, hotspot, and difficulty of current research. In this study, to achieve high precision quantitative analysis under complex detection conditions, utilizing the fusion of multi-dimensional plasma information and the integration of physical models and algorithmic models, a spectral bias-error stepwise correction method of plasma image-spectrum fusion based on deep learning (SBESC-PISF) was proposed. In this method, based on the statistical properties of LIBS spectra, the actual obtained spectra were decomposed into three parts: the ideal spectral intensity related only to the element concentration, and the spectral bias and spectral error caused by the fluctuation of complex high-dimensional plasma parameters. Further, the deep learning methods were used to fully excavate all the effective features in the plasma images and spectra to invert the complex high-dimensional plasma parameters according to the physical models. Finally, the estimation models of spectral bias and spectral error were established based on these features, to realize the high-precision correction of spectral intensity. To verify the feasibility of SBESC-PISF, the spectra of aluminum alloy samples obtained under three complex detection conditions were used for analysis. Under the experimental condition of laser energy fluctuation, after correction by SBESC-PISF, R2 of the three calibration curves was all increased to 0.999, RMSE and STD of the validation set (RMSEV, STDV) were reduced by 55.246 % and 50.167 %, respectively. Under the experimental condition of defocusing amount fluctuation, R2 was also all increased to 0.999, RMSEV and STDV were decreased by 58.201 % and 51.006 %, respectively. When the laser energy and defocusing amount fluctuate simultaneously, R2 was increased to 0.999, 0.996 and 0.988, RMSEV and STDV were reduced by 58.776 % and 54.397 %, respectively. These experimental results demonstrate that the spectral fluctuation correction of SBESC-PISF under complex detection conditions is effective and has wide applicability.

Prediction Models for Perioperative Blood Transfusion in Patients Undergoing Gynecologic Surgery: A Systematic Review.

This systematic review aimed to evaluate prediction models for perioperative blood transfusion in patients undergoing gynecologic surgery. Given the inherent risks associated with blood transfusion and the critical need for accurate prediction, this study identified and assessed models based on their development, validation, and predictive performance. The review included five studies encompassing various surgical procedures and approaches. Predicting factors commonly used across these models included preoperative hematocrit, race, surgical route, and uterine fibroid characteristics. However, the review highlighted significant variability in the definition of perioperative periods, a lack of standardization in transfusion criteria, and a high risk of bias in most models due to methodological issues, such as a low number of events per variable, inappropriate handling of continuous and categorical predictors, inappropriate handling of missing data, improper methods of predictor selection, inappropriate measurement methods for model performance, and inadequate evaluations of model overfitting and optimism in model performance. Despite some models demonstrating good discrimination and calibration, the overall quality and external validation of these models were limited. Consequently, there is a clear need for more robust and externally validated models to improve clinical decision-making and patient outcomes in gynecologic surgery. Future research should focus on refining these models, incorporating rigorous validation, and adhering to standardized reporting practices.

The Effects of the 2020 BLM Protests on Racial Bias in the United States.

The 2020 Black Lives Matter (BLM) protests in response to the murder of George Floyd highlighted the lingering structural inequalities faced by Black people in the United States. In the present research, we investigated whether these protests led to reduced implicit and explicit racial bias among White U.S. Americans. Combining data from Project Implicit, Armed Conflict Location Event Data Project (ACLED), Google Trends, and the American Community survey, we observed rapid drops in implicit and explicit measures of racial bias after the onset of the protests. However, both types of racial bias slowly increased again over time as (attention to) BLM faded. We use directed acyclic graphs to show under which assumptions causal inferences are warranted. We discuss our results in light of situational models of bias, their implications for protest movements, and raise questions about when and how social norms play a role in large-scale attitude change.

Assessing South Indian Ocean tropical cyclone characteristics in HighResMIP simulations

AbstractSeveral damaging tropical cyclones (TCs) have occurred recently over the South Indian Ocean (SIO) region, causing enormous social and economic losses. Yet, while many studies have examined SIO TC characteristics using observations and reanalysis, only a few have assessed these characteristics specifically for this region in climate models, and fewer have investigated their projections under climate change. Here we do this for a historical (1980–2010) and future (2020–2050) period, using multimodel simulations from the High Resolution Model Intercomparison Project, as well as examine biases in the historical period relative to a reanalysis (ERA5). The models have horizontal resolutions of 25–50 km, which has enabled an improved ability to represent tropical cyclones globally in previous studies. TempestExtremes software is employed to detect tropical storm and cyclone tracks. In cases where TempestExtremes cannot be applied due to a lack of requisite variables in a dataset, we instead examine extreme wind speeds in that dataset. For the historical period, we find considerable variation in model biases compared to ERA5, which itself exhibits realistic spatial patterns of tracks and their monthly distribution. Models do at least agree on positive biases in track frequency east of Madagascar and somewhat in the Mozambique Channel. However, the models and ERA5 only produce Category 3 tropical cyclones at best. Wind speeds for 25 km resolution models have much larger positive biases than for 50 km ones, suggesting the former can simulate even higher‐category tropical cyclones. Considerable intermodel variation is also found in track changes between the future and historical periods. No systematic intercategory pattern of change exists, and low signal‐to‐noise may obscure any such patterns in the limited timespan of available data. Thus, no meaningful conclusions can be drawn regarding changes in track intensity. Nevertheless, track frequency broadly decreases across models for the region, as does accumulated cyclone energy. An east‐to‐west shift in track location from east of Madagascar toward the Mozambique Channel is also implied by track frequency and wind speed changes. Our findings provide information to potentially improve storm resiliency in this vulnerable region.

Regional aggregation bias in the global computable general equilibrium model: Issues in assessing the economic impact of climate change and trade liberalization

This study examined the bias introduced by regional aggregation (RA) when assessing the effect of climate change and trade liberalization through computational experiments. The analysis used global computable general equilibrium models with different pre-aggregation levels, based on the GTAP10 database. The bias was quantified by comparing the effects between large-scale and aggregated models. The results revealed that RA bias comprised over 35% of estimated effects related to climate change and global trade liberalization; it reversed the change direction in the simulation and deviated changes beyond the variation range observed in the sensitivity analysis; and it increased in developing countries where the proportions of imports were high. Much RA bias is artificially created by aggregating countries with different initial shocks, which then spreads to other countries through a pseudo-trade structure. Therefore, adopting a disaggregated global model is crucial to improve the accuracy of climate change assessments with diverse shocks worldwide.

Assessment of the Southern Ocean Sea Surface Temperature Biases in CMIP5 and CMIP6 Models

Assessment of the Southern Ocean Sea Surface Temperature Biases in CMIP5 and CMIP6 Models

Unmasking and quantifying racial bias of large language models in medical report generation

BackgroundLarge language models like GPT-3.5-turbo and GPT-4 hold promise for healthcare professionals, but they may inadvertently inherit biases during their training, potentially affecting their utility in medical applications. Despite few attempts in the past, the precise impact and extent of these biases remain uncertain.MethodsWe use LLMs to generate responses that predict hospitalization, cost and mortality based on real patient cases. We manually examine the generated responses to identify biases.ResultsWe find that these models tend to project higher costs and longer hospitalizations for white populations and exhibit optimistic views in challenging medical scenarios with much higher survival rates. These biases, which mirror real-world healthcare disparities, are evident in the generation of patient backgrounds, the association of specific diseases with certain racial and ethnic groups, and disparities in treatment recommendations, etc.ConclusionsOur findings underscore the critical need for future research to address and mitigate biases in language models, especially in critical healthcare applications, to ensure fair and accurate outcomes for all patients.

Steric Sea Level Rise and Relationships with Model Drift and Water Mass Representation in GFDL CM4 and ESM4

Abstract Density-driven steric seawater changes are a leading-order contributor to global mean sea level rise. However, inter-model differences in the magnitude and spatial patterns of steric sea level rise exist at regional scales and often emerge during the spin-up and pre-industrial control integrations of climate models. Steric sea level results from an eddy-permitting climate model, GFDL-CM4, are compared with a lower resolution counterpart, GFDL-ESM4. The results from both models are examined through basin-scale heat budgets and watermass analysis, and we compare the patterns of ocean heat uptake, redistribution, and sea level differ in ocean-only (i.e. OMIP) and coupled climate configurations. After correcting for model drift, both GFDL-CM4 and GFDL-ESM4 simulate nearly equivalent ocean heat content change and global sea level rise during the historical period. However, the GFDL-CM4 model exhibits as much as a 40% increase in surface ocean heat uptake in the Southern Ocean and subsequent increases in horizontal export to other ocean basins after bias correction. The results suggest regional differences in the processes governing Southern Ocean heat export, such as the formation of AAIW, SPMW, and gyre transport between the two models, and that sea level changes in these models cannot be fully bias-corrected. Since the process-level differences between the two models are evident in the preindustrial control simulations of both models, these results suggest that the control simulations are important for identifying and correcting sea-level related model biases.

Does Applying Subsampling in Quantile Mapping Affect the Climate Change Signal?

Bias in regional climate model (RCM) data makes bias correction (BC) a necessary pre-processing step in climate change impact studies. Among a variety of different BC methods, quantile mapping (QM) is a popular and powerful BC method. Studies have shown that QM may be vulnerable to reductions in calibration sample size. The question is whether this also affects the climate change signal (CCS) of the RCM data. We applied four different QM methods without subsampling and with three different subsampling timescales to an ensemble of seven climate projections. BC generally improved the RCM data relative to observations. However, the CCS was significantly modified by the BC for certain combinations of QM method and subsampling timescale. In conclusion, QM improves the RCM data that are fundamental for climate change impact studies, but the optimal subsampling timescale strongly depends on the chosen QM method.

Optimization of renewable energy hybrid power plant in the state of Karnataka, India

This paper proposes a methodology on reliability and cost minimization of the electricity produced by Renewable energy sources. The approach is to use Hybrid Renewable Energy Systems, which involves integrating various renewable sources. The objective of this study is to address these challenges by HRES method in the state of Karnataka. In Karnataka, there has been large scale curtailment of renewable energy generation due to the various structural issues. This study would give a solution to reliability and cost related issues faced by this state. For the study a Mathematical Model has been developed using Feed Forward Neural Network (FNN) and Chemical Reaction Optimization algorithms. Chemical reaction Optimization model is chosen to remove biases of the Artificial Neural Network Model of FNN. This model will be used to simulate interaction between Hybrid Renewable Energy Systems. The result of the study will provide a power system with high & optimized renewable energy output in the state of Karnataka, India which will be reliable and cost effective.