Bias Elimination Research Articles

A review of research on training responsible large language models

Abstract. In recent years, there has been a growing acceptance of large language models (LLM) as a mainstream method in the field of natural language processing. Consequently, numerous studies have been conducted on this topic. Training responsible Large Language Models have become a prominent subject of research in the past few years. This type of research mainly focuses on the examination of bias, morality and other aspects of LLM. There are certain similarities in the methodologies employed in those studies. This article presents a comprehensive overview of numerous recent investigations, analyzing and categorizing the methodologies employed in these studies, and offering a literature review. This review examines the three perspectives of LLM bias data set construction, bias detection and bias elimination It provides a comparative analysis of the advantages and disadvantages of different methods. After completing relevant evaluations, a comprehensive examination of the research on training responsible LLM is conducted and potential future research directions are proposed in this article.

TransDiffSeg: Transformer-Based Conditional Diffusion Segmentation Model for Abdominal Multi-Objective.

In the domain of medical image segmentation, traditional diffusion probabilistic models are hindered by local inductive biases stemming from convolutional operations, constraining their ability to model long-term dependencies and leading to inaccurate mask generation. Conversely, Transformer offers a remedy by obviating the local inductive biases inherent in convolutional operations, thereby enhancing segmentation precision. Currently, the integration of Transformer and convolution operations mainly occurs in two forms: nesting and stacking. However, both methods address the bias elimination at a relatively large granularity, failing to fully leverage the advantages of both approaches. To address this, this paper proposes a conditional diffusion segmentation model named TransDiffSeg, which combines Transformer with convolution operations from traditional diffusion models in a parallel manner. This approach eliminates the accumulated local inductive bias of convolution operations at a finer granularity within each layer. Additionally, an adaptive feature fusion block is employed to merge conditional semantic features and noise features, enhancing global semantic information and reducing the Transformer's sensitivity to noise features. To validate the impact of granularity in bias elimination on performance and the impact of Transformer in alleviating the accumulated local inductive biases of convolutional operations in diffusion probabilistic models, experiments are conducted on the AMOS22 dataset and BTCV dataset. Experimental results demonstrate that eliminating local inductive bias at a finer granularity significantly improves the segmentation performance of diffusion probabilistic models. Furthermore, the results confirm that the finer the granularity of bias elimination, the better the segmentation performance.

Empowering nurses to champion Health equity & BE FAIR: Bias elimination for fair and responsible AI in healthcare.

The concept of health equity by design encompasses a multifaceted approach that integrates actions aimed at eliminating biased, unjust, and correctable differences among groups of people as a fundamental element in the design of algorithms. As algorithmic tools are increasingly integrated into clinical practice at multiple levels, nurses are uniquely positioned to address challenges posed by the historical marginalization of minority groups and its intersections with the use of "big data" in healthcare settings; however, a coherent framework is needed to ensure that nurses receive appropriate training in these domains and are equipped to act effectively. We introduce the Bias Elimination for Fair AI in Healthcare (BE FAIR) framework, a comprehensive strategic approach that incorporates principles of health equity by design, for nurses to employ when seeking to mitigate bias and prevent discriminatory practices arising from the use of clinical algorithms in healthcare. By using examples from a "real-world" AI governance framework, we aim to initiate a wider discourse on equipping nurses with the skills needed to champion the BE FAIR initiative. Drawing on principles recently articulated by the Office of the National Coordinator for Health Information Technology, we conducted a critical examination of the concept of health equity by design. We also reviewed recent literature describing the risks of artificial intelligence (AI) technologies in healthcare as well as their potential for advancing health equity. Building on this context, we describe the BE FAIR framework, which has the potential to enable nurses to take a leadership role within health systems by implementing a governance structure to oversee the fairness and quality of clinical algorithms. We then examine leading frameworks for promoting health equity to inform the operationalization of BE FAIR within a local AI governance framework. The application of the BE FAIR framework within the context of a working governance system for clinical AI technologies demonstrates how nurses can leverage their expertise to support the development and deployment of clinical algorithms, mitigating risks such as bias and promoting ethical, high-quality care powered by big data and AI technologies. As health systems learn how well-intentioned clinical algorithms can potentially perpetuate health disparities, we have an opportunity and an obligation to do better. New efforts empowering nurses to advocate for BE FAIR, involving them in AI governance, data collection methods, and the evaluation of tools intended to reduce bias, mark important steps in achieving equitable healthcare for all.

MOTC: Abdominal Multi-objective Segmentation Model with Parallel Fusion of Global and Local Information.

Convolutional Neural Networks have been widely applied in medical image segmentation. However, the existence of local inductive bias in convolutional operations restricts the modeling of long-term dependencies. The introduction of Transformer enables the modeling of long-term dependencies and partially eliminates the local inductive bias in convolutional operations, thereby improving the accuracy of tasks such as segmentation and classification. Researchers have proposed various hybrid structures combining Transformer and Convolutional Neural Networks. One strategy is to stack Transformer blocks and convolutional blocks to concentrate on eliminating the accumulated local bias of convolutional operations. Another strategy is to nest convolutional blocks and Transformer blocks to eliminate bias within each nested block. However, due to the granularity of bias elimination operations, these two strategies cannot fully exploit the potential of Transformer. In this paper, a parallel hybrid model is proposed for segmentation, which includes a Transformer branch and a Convolutional Neural Network branch in encoder. After parallel feature extraction, inter-layer information fusion and exchange of complementary information are performed between the two branches, simultaneously extracting local and global features while eliminating the local bias generated by convolutional operations within the current layer. A pure convolutional operation is used in decoder to obtain final segmentation results. To validate the impact of the granularity of bias elimination operations on the effectiveness of local bias elimination, the experiments in this paper were conducted on Flare21 dataset and Amos22 dataset. The average Dice coefficient reached 92.65% on Flare21 dataset, and 91.61% on Amos22 dataset, surpassing comparative methods. The experimental results demonstrate that smaller granularity of bias elimination operations leads to better performance.

Eliminating Contextual Prior Bias for Semantic Image Editing via Dual-Cycle Diffusion

The recent success of text-to-image generation diffusion models has also revolutionized semantic image editing, enabling the manipulation of images based on query/target texts. Despite these advancements, a significant challenge lies in the potential introduction of contextual prior bias in pre-trained models during image editing, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g</i> ., making unexpected modifications to inappropriate regions. To address this issue, we present a novel approach called Dual-Cycle Diffusion, which generates an unbiased mask to guide image editing. The proposed model incorporates a Bias Elimination Cycle that consists of both a forward path and an inverted path, each featuring a Structural Consistency Cycle to ensure the preservation of image content during the editing process. The forward path utilizes the pre-trained model to produce the edited image, while the inverted path converts the result back to the source image. The unbiased mask is generated by comparing differences between the processed source image and the edited image to ensure that both conform to the same distribution. Our experiments demonstrate the effectiveness of the proposed method, as it significantly improves the D-CLIP score from 0.272 to 0.283. The code will be available at https://github.com/JohnDreamer/DualCycleDiffsion.

Improved literature recommendation system through the fusion mode of conformity bias elimination and keyword preference

Authors are often influenced by historical experience and personal habits when selecting keywords in composing papers, and tend to choose specific keywords, resulting in a phenomenon of conformity bias. In order to improve the accuracy of literature recommendation services, it is crucial to effectively identify and eliminate the adverse effects of conformity bias. Based on the definition of a measurement for the degree of impact of conformity bias using conformity width, this paper designs an effective method for eliminating conformity bias based on one-class implicit access information. At the same time, this method further utilizes the author's own keyword preference information based on explicit weight of keywords and achieves the comprehensive fusion of various effective user access information through the design of a model. The experimental results validated the phenomenon of conformity bias in the composing process of scholars, and comprehensively analyzed and compared the characteristics and advantages of the algorithm proposed in this paper. The code is available at https://github.com/EdwardSig/CBEWC_ESWA.

An Improved Dual Second-Order Generalized Integrator Phased-Locked Loop Strategy for an Inverter of Flexible High-Voltage Direct Current Transmission Systems under Nonideal Grid Conditions

High-voltage flexible power systems, with their intrinsic characteristics, play an increasingly important role in electronic power systems. Synchronization between the inverter and the grid needs to be achieved by a phase-locked loop (PLL), the performance of which determines the quality of power transmission. This paper proposes a PLL adapted to extremely harsh grid conditions. Firstly, the traditional synchronous reference frame PLL and the dual second-order generalized integrator (DSOGI-PLL) are analyzed, and the errors in phase-locking and the shortcomings of these two methods in the presence of DC components in the grid are pointed out. Secondly, based on the harmonic grid voltage, a repetitive control internal model is introduced by DSOGI to realize the real-time tracking and regulation of the harmonic signals in order to suppress the harmonic voltage disturbance. In addition, a DC bias elimination and frequency adaptive method is proposed to solve the problems of DC bias and grid voltage frequency fluctuation in order to achieve adaptive tracking of the grid phase. Finally, the superiority of the proposed strategy is verified through simulations and experiments.

A nationwide, prospective, cohort study on exogenous oxytocin and delays in early child development: the Japan environment and children's study.

Although it remains debatable, exogenous oxytocin, commonly used for labour induction and augmentation, reportedly increases risks of neurodevelopment delay, attention-deficit/hyperactivity disorder, and autism spectrum disorder among children prenatally exposed to exogenous oxytocin. However, only few studies have objectively examined exogenous oxytocin's impact on early childhood development through scoring evaluations. This study investigated the association between exogenous oxytocin exposure and neurodevelopment in 3-year-old children using the Ages and Stages Questionnaires, Third Edition. In this nationwide prospective cohort study, we extracted data from 104,062 foetal records regarding exogenous oxytocin use during labour from the Japan Environment and Children's Study. Participants completed questionnaires throughout the pregnancy and postpartum periods. Outcomes comprised the developmental status less than each cut-off value for the five domains of the Ages and Stages Questionnaire, Third Edition. We conducted multivariable logistic regression analyses on the data of 55,400 children after controlling for confounders. Among the 55,400 included women, 19.0% (n = 10,506) used exogenous oxytocin during labour and 81.0% (n = 44,894) did not. Children exposed to exogenous oxytocin showed no significantly increased risk of developmental delay in any domain (communication: odds ratio [OR] 1.04, 95% confidence interval [CI] 0.92-1.16; gross motor: OR 0.97, 95% CI 0.87-1.08; fine motor: OR 1.00, 95% CI 0.92-1.09; problem-solving: OR 1.02, 95% CI 0.94-1.11; personal-social: OR 0.91, 95% CI 0.80-1.03). Conclusion: Exogenous oxytocin for labour induction did not adversely affect early childhood development. Further studies accounting for the degree of exogenous oxytocin exposure are required to confirm these results. What is Known: • In developed countries, labour is induced in 20-25% of all pregnancies, for which oxytocin is commonly used. • Studies have associated risks of neurodevelopment delay, attention-deficit/hyperactivity disorder, and autism spectrum disorder with exposure to exogenous oxytocin. What is New: • Evaluation with the Ages and Stages Questionnaire, Third Edition, revealed that exogenous oxytocin use did not adversely affect early childhood development. • This prospective study reinforced the lack of evidence of an association between exogenous oxytocin use and early childhood development after adjustment for confounding and rigorous bias elimination.

DC Bias Suppression in Static Power Converter During Catenary Fault Handling for Railway Flexible AC Traction Power System

Flexible AC traction power supply system (FTPSS) has been a potential candidate for high-speed and heavy-haul railways. The catenary fault is a crucial issue in the FTPSS. Since the static power converters (SPCs) become the main source, SPC is required to switch between the voltage-frequency and the fault-current control modes, in order to achieve fault detection and rapid system recovery. However, inappropriate switching method may lead to violent change of instantaneous voltage. Therefore, serious DC bias of SPC output transformer may occur and excite severe overcurrent. It results in the tripping of SPCs and the failure of system recovery, and severely undermine the reliability of FTPSS. To address this issue, this paper analyzes and reveals the mechanism of this DC bias phenomenon. The voltage-time product imbalance during the mode switching process is the main cause of DC bias, and it is related to the voltage phase at the switching instants. Accordingly, a combined method of voltage waveform splicing and DC bias elimination control is proposed. Overcurrent can be avoided, and smooth control-mode switching and rapid system recovery can be achieved. Simulation and experimental results verify the validation of the proposed method.

Three Satellites Dynamic Switching Range Integrated Navigation and Positioning Algorithm with Clock Bias Cancellation and Altimeter Assistance

Challenging environments such as cities, canyons, and forests have become key factors affecting navigation stability. When users pass through intricate overpasses and winding road sections, due to the fluctuation of the geoid, there will be a large fluctuation problem in the elevation measurement error of the user’s receiver. In addition, even if the low Earth orbit (LEO) constellation has thousands of satellites, there will be no technical problems in regard to destroying LEO satellites with existing technology in extreme situations such as warfare and in challenging environments such as dense forests, canyons, and ravines, where three or fewer visible satellites is a foreseeable scenario. To solve the problem of providing location services in such challenging environments, first, we analyze the relationship between temperature and atmospheric pressure and altitude; and then, based on this, we propose an initialization correction method for elevation measurements. Next, based on the broadband LEO constellation, we give an integrated navigation and positioning scheme with the assistance of both a clock bias elimination system and an altimeter. Finally, the proposed scheme is simulated and verified. The experimental results show that the dynamic switching of LEO satellites, combined with the assistance of the altimeter, can effectively improve the stability and positioning accuracy of navigation and positioning and can suppress the large navigation errors caused by the long switching time without the assistance of the altimeter. This allows the switching time to be extended; thus, it can be used as a technical reference solution for integrated communication and navigation (ICN) in the future.

Bias in nutrition-health associations is not eliminated by excluding extreme reporters in empirical or simulation studies.

Self-reported nutrition intake (NI) data are prone to reporting bias that may induce bias in estimands in nutrition studies; however, they are used anyway due to high feasibility. We examined whether applying Goldberg cutoffs to remove 'implausible' self-reported NI could reliably reduce bias compared to biomarkers for energy, sodium, potassium, and protein. Using the Interactive Diet and Activity Tracking in the American Association of Retired Persons (IDATA) data, significant bias in mean NI was removed with Goldberg cutoffs (120 among 303 participants excluded). Associations between NI and health outcomes (weight, waist circumference, heart rate, systolic/diastolic blood pressure, and VO2 max) were estimated, but sample size was insufficient to evaluate bias reductions. We therefore simulated data based on IDATA. Significant bias in simulated associations using self-reported NI was reduced but not completely eliminated by Goldberg cutoffs in 14 of 24 nutrition-outcome pairs; bias was not reduced for the remaining 10 cases. Also, 95% coverage probabilities were improved by applying Goldberg cutoffs in most cases but underperformed compared with biomarker data. Although Goldberg cutoffs may achieve bias elimination in estimating mean NI, bias in estimates of associations between NI and outcomes will not necessarily be reduced or eliminated after application of Goldberg cutoffs. Whether one uses Goldberg cutoffs should therefore be decided based on research purposes and not general rules.

Need Analysis of Indian Critical Health Care Delivery in Government Sectors and Its Impact on the General Public: A Time to Revamp Public Health Care Infrastructure.

Poverty is directly linked to public health care delivery in many ways and dimensions. Every aspect of the human sphere is preplanned, but a health crisis is the only emergency which pushes humanity into severe economic stress. Therefore, every nation aims to safeguard its citizens from a health crisis. In this aspect, India needs to improve its public health infrastructure in order to protect its citizens and save them from poverty. (1) To assess the current pitfalls in public critical health care delivery, (2) to analyze whether the health care delivery matches the requirements of its population in every state, (3) to produce solutions and guidelines to overcome the stress in this priority area. Data regarding the critical care workforce, which includes critical care doctors and nurses, were taken from official websites and other sources. Critical care infrastructure data were retrieved from the Internet sources. Data were validated by consulting state government sources and cross-checked for bias elimination. The data were analyzed using the "Statistical Package for Social Sciences" software version 20, and were presented using descriptive statistics. There is a 1:10 percentage of deficit in the case of critical care workforce and infrastructure when compared with its need analysis. Critical care medicine specialists are in 1:75 when compared to other specialties. Overall, the public sector critical care needs a total boost through out of box solutions. According to the Stockholm International Peace Research Institute (SIPRI), India spent the third most on defense in the world in 2021. India spent 76.6 billion dollars on its military in 2021, up 33% from 2012 and 0.9% from 2020. However, since India is considered a fast-growing economy, there is still a huge disparity in critical care. Without resetting critical health care, India cannot grow in welfare indices even if it is among the top gross domestic product (GDP) countries. Prabu D, Gousalya V, Rajmohan M, Dinesh MD, Bharathwaj VV, Sindhu R, et al. Need Analysis of Indian Critical Health Care Delivery in Government Sectors and its Impact on the General Public: A Time to Revamp Public Health Care Infrastructure. Indian J Crit Care Med 2023;27(4):237-245.

Subjectivity Bias Elimination in Employee Appraisal Process – Role of People Analytics

Purpose: This research serves the purpose of dealing with different types of Human Subjective Biases in Performance Management System by interview-based analysis and content analysis depending upon inputs from renowned HR professionals and then to determine if HR Analytics can help the HR practitioners in objective-decision making rather than intuition in PMS.Design/Methodology/Approach: Focused-Group-Interview approach is followed here to acquire in-depth information about Subjectivity Bias Elimination in Employee Appraisal Process and how the same can be mitigated with the help of People Analytics. Findings: This research will reveal the truth of erroneous subjective judgments based on instincts and how HR Analytics can help the HR practitioners in objective decision making rather than intuition in PMS. Originality/ Value: This research identifies factors leading to Subjectivity Bias in Employee Appraisal Process. It suggests updated criteria for HR professionals to mitigate subjectivity bias in Employee Appraisal Process with help of People Analytics.

Enhancing the reliability and accuracy of data-driven dynamic soft sensor based on selective dynamic partial least squares models

Data-driven soft sensors have been widely applied to a broad range of process industries for virtually sensing difficult-to-measure but of-great-concern variables. However, it is still nontrivial to develop dynamic soft sensors with satisfactory performance. A crucial obstacle lies in strong dependence on quality and representativity of collected data. In other words, the predictive accuracy of the developed soft sensor could be quite sensitive to both offline training data and online unseen data, leading to unreliable and degraded generalization performance. In order to deal with such a troublesome issue, this paper proposes a framework for developing reliable dynamic soft sensor called selective dynamic partial least squares (SDPLS). The SDPLS consists of two-stage operations. At the offline stage, aided by intelligent optimization algorithm a model library is established through constructing various DPLS models, each of which accounts for certain ‘mode’. At the online stage, adaptive online model updating for adapting to the current working condition is carried out based on evaluating the performance of the stored individual models, where a correction scheme is also developed for bias elimination. Extensive case studies have been conducted based on a numerical example and two real-life industrial processes, and the results efficaciously demonstrate the effectiveness and promising application prospects of the proposed schemes.

Consistency analysis and bias elimination of the Instrumental-Variable-based State Variable Filter method

In this paper, we analyse the consistency of the Instrumental-Variable-based State Variable Filter (IVSVF) estimator by taking into account the intersample behaviour of the input and output signals. It is found that when only sampled input and output data are available for estimation, the IVSVF estimator is not consistent for any fixed sampling period due to the interpolation error that arises from constructing the filtered output. A Bias-Eliminated IVSVF (BEIVSVF) estimator is then proposed and shown to be consistent. The theoretical results developed in the paper are also discussed from a practical standpoint. Simulations are performed to verify the performance of the proposed method as well as to support the theoretical results.

A Fluxgate-Based Current Sensor for DC Bias Elimination in a Dual Active Bridge Converter

A concern with the isolation transformer in a dual active bridge (DAB) dc–dc converter is the dc bias in magnetization. This article proposes a fluxgate-based current sensor to measure the dc component mixed with a large, high-frequency ac current. Compared with a commercial Hall effect current sensor, the proposed sensor significantly reduces measurement error. This article presents the working principle and design considerations. A prototype is demonstrated for dc bias elimination control in a DAB converter.

A New Approach for Explicit Distributional Semantic Vector Representation Based on Dependency Distance and Bias Elimination

We consider the problem of the semantic representation of words based on explicit basis word vectors. We focus on explicit semantic vectors in which each basis vector is equivalent to a unique word. In this paper, we propose a new approach to represent semantic vectors explicitly. The approach consists of three steps: First, we introduce two new factors called word similarity (WS) and the number of zeroes (NZ) for choosing informative context words. Second, we propose a dependency distance method by incorporating the syntactic relations of the words in the sentences. This means that the method uses not only semantic information but also syntactic one. Third, we propose two shifting methods based on WF and WS factors to eliminate the bias of the PPMI matrix. In this work, we use the ukWaC corpus to select context words and compute co-occurrence matrices. Then, we examine the results on MEN, RG-65, SimLex-999, and WordSim353 test sets. We report the performance in comparison with a Baseline in which, the word frequency has been used. The Baseline method uses a window to fix the influence of the neighboring words. Our approach improves the Spearman correlation coefficient for MEN, RG-65, SimLex-999, and WordSim353 datasets by 3.87%, 10.25%, 5.04%, and 5.24%, respectively. So, the experiment results show that the proposed context words selection, incorporating dependency distance, and using the WF-based shifting method significantly improve the performance.

Discriminative Extreme Learning Machine with Cross‐Domain Mean Approximation for Unsupervised Domain Adaptation

Extreme Learning Machine (ELM) is widely used in various fields because of its fast training and high accuracy. However, it does not primarily work well for Domain Adaptation (DA) in which there are many annotated data from auxiliary domain and few even no annotated data in target domain. In this paper, we propose a new variant of ELM called Discriminative Extreme Learning Machine with Cross‐Domain Mean Approximation (DELM‐CDMA) for unsupervised domain adaptation. It introduces Cross‐Domain Mean Approximation (CDMA) into the hidden layer of ELM to reduce distribution discrepancy between domains for domain bias elimination, which is conducive to train a high accuracy ELM on annotated data from auxiliary domains for target tasks. Linear Discriminative Analysis (LDA) is also adopted to improve the discrimination of learned model and obtain higher accuracy. Moreover, we further provide a Discriminative Kernel Extreme Learning Machine with Cross‐Domain Mean Approximation (DKELM‐CDMA) as the kernelization extension of DELM‐CDMA. Some experiments are performed to investigate the proposed approach, and the result shows that DELM‐CDMA and DKELM‐CDMA could effectively extend ELM suitable for domain adaptation and outperform ELM and many other domain adaptation approaches.

Aspects of Regional and Worldwide Mineral Resource Prediction

The purpose of this contribution is to highlight four topics of regional and worldwide mineral resource prediction: (1) use of the jackknife for bias elimination in regional mineral potential assessments; (2) estimating total amounts of metal from mineral potential maps; (3) fractal/multifractal modeling of mineral deposit density data in permissive areas; and (4) worldwide and large-areas metal size-frequency distribution modeling. The techniques described in this paper remain tentative because they have not been widely researched and applied in mineral potential studies. Although most of the content of this paper has previously been published, several perspectives for further research are suggested.

Single-Satellite Integrated Navigation Algorithm Based on Broadband LEO Constellation Communication Links

With the rapid development of satellite technology and the need to satisfy the increasing demand for location-based services, in challenging environments such as indoors, forests, and canyons, there is an urgent need to improve the position accuracy in these environments. However, traditional algorithms obtain the position solution through time redundancy in exchange for spatial redundancy, and they require continuous observations that cannot satisfy the real-time location services. In addition, they must also consider the clock bias between the satellite and receiver. Therefore, in this paper, we provide a single-satellite integrated navigation algorithm based on the elimination of clock bias for broadband low earth orbit (LEO) satellite communication links. First, we derive the principle of LEO satellite communication link clock bias elimination; then, we give the principle and process of the algorithm. Next, we model and analyze the error of the system. Subsequently, based on the unscented Kalman filter (UKF), we model the state vector and observation vector of our algorithm and give the state and observation equations. Finally, for different scenarios, we conduct qualitative and quantitative analysis through simulations, and the results show that, whether in an altimeter scenario or non-altimeter scenario, the performance indicators of our algorithm are significantly better than the inertial navigation system (INS), which can effectively overcome the divergence problem of INS; compared with the medium earth orbit (MEO) constellation, the navigation trajectory under the LEO constellation is closer to the real trajectory of the aircraft; and compared with the traditional algorithm, the accuracy of each item is improved by more than 95%. These results show that our algorithm not only significantly improves the position error, but also effectively suppresses the divergence of INS. The algorithm is more robust and can satisfy the requirements of cm-level real-time location services in challenging environments.