Intrinsic Bias Research Articles

Microstructural Underpinnings of Giant Intrinsic Exchange Bias in Epitaxial NiCo2O4 Thin Films

AbstractUnderstanding intrinsic exchange bias in nominally single‐component ferromagnetic or ferrimagnetic materials is crucial for simplifying related device architectures. However, the mechanisms behind this phenomenon and its tunability remain elusive, which hinders the efforts to achieve unidirectional magnetization for widespread applications. Inspired by the high tunability of ferrimagnetic inverse spinel NiCo2O4, the origin of intrinsic exchange bias in NiCo2O4 (111) films deposited on Al2O3 (0001) substrates are investigated. The comprehensive characterizations, including electron diffraction, X‐ray reflectometry and spectroscopy, and polarized neutron reflectometry, reveal that intrinsic exchange bias in NiCo2O4 (111)/Al2O3 (0001) arises from a reconstructed antiferromagnetic rock‐salt NixCo1‐xO layer at the interface between the film and the substrate due to a significant structural mismatch. Remarkably, by engineering the interfacial structure under optimal growth conditions, it can achieve exchange bias larger than coercivity, leading to unidirectional magnetization. Such giant intrinsic exchange bias can be utilized for realistic device applications. This work establishes a new material platform based on NiCo2O4, an emergent spintronics material, to study tunable interfacial magnetic and spintronic properties.

Unboxing Occupational Bias: Debiasing LLMs with U.S. Labor Data

Large Language Models (LLMs) are prone to inheriting and amplifying societal biases embedded within their training data, potentially reinforcing harmful stereotypes related to gender, occupation, and other sensitive categories. This issue becomes particularly problematic as biased LLMs can have far-reaching consequences, leading to unfair practices and exacerbating social inequalities across various domains, such as recruitment, online content moderation, or even the criminal justice system. Although prior research has focused on detecting bias in LLMs using specialized datasets designed to highlight intrinsic biases, there has been a notable lack of investigation into how these findings correlate with authoritative datasets, such as those from the U.S. National Bureau of Labor Statistics (NBLS). To address this gap, we conduct empirical research that evaluates LLMs in a “bias-out-of-the-box” setting, analyzing how the generated outputs compare with the distributions found in NBLS data. Furthermore, we propose a straightforward yet effective debiasing mechanism that directly incorporates NBLS instances to mitigate bias within LLMs. Our study spans seven different LLMs, including instructable, base, and mixture-of-expert models, and reveals significant levels of bias that are often overlooked by existing bias detection techniques. Importantly, our debiasing method, which does not rely on external datasets, demonstrates a substantial reduction in bias scores, highlighting the efficacy of our approach in creating fairer and more reliable LLMs.

Are food and beverage purchases reflective of dietary intake? Validity of supermarket purchases as indicator of diet quality in the Supreme Nudge Trial.

Dietary intake assessment is often complicated by intrinsic bias. This study investigated whether food purchase data could constitute a valid indication of dietary intake, by evaluating the extent to which diet quality as measured by supermarket food purchases is correlated with diet quality as measured by reported dietary intake. We used data from the Supreme Nudge cluster-randomised controlled supermarket trial (n=227). Data were collected at baseline from supermarket purchases (loyalty cards) and a dietary questionnaire (short 40-item food frequency questionnaire (FFQ)) to compute two scores reflecting diet quality from purchasing data (purchased diet quality) and FFQs (consumed diet quality). Both scores constituted of 13 food groups and could theoretically range between 0 (low diet quality) to 130 (high diet quality). The relationship between purchased diet quality and consumed diet quality was assessed using correlation coefficients, and the Bland-Altman limits-of-agreement method. Multiple linear regression was fitted between purchased diet quality and consumed diet quality, adjusted for age, sex, waist circumference, educational level, and household size. Consumed and purchased diet qualities were modestly positively correlated (Pearson's ρ = 0·31, 95% CI: 0·18 - 0·42). A positive association from linear regression was found after confouding adjustments (βbaseline = 0·22, 95%CI: 0·10 - 0·34). Purchased diet quality was systematically lower than the consumed diet quality. This study found that diet quality as measured by supermarket purchases provided a reasonable indication of diet quality as reported by short-FFQs, albeit modest.

Updated CP*Trends: An Online Tool to Compare Cohort and Period Trends across Cancer Sites.

CP*Trends is a widely used SEER website used to explore temporal effects of period and cohort on cancer incidence and mortality. It provides a graphical display of smoothed rates, and a C-P Score that helps to assess the magnitude of the effect of cohort and period. This update provides results for African Americans and Whites. The C-P Score has an intrinsic bias favoring cohort because there are many more cohorts than periods. An adjusted C-P Score removes some of this advantage. Bootstrap confidence intervals are given, which allow one to see the effects of different sample sizes on the model results. Finally, users may control window size used in the smoothing algorithm, which helps to avoid over smoothing or masking of trends. The method is illustrated using data on cervical cancer incidence trends for African Americans and Whites, 1975-2018. Rates are higher for African Americans, and both races have contributions for cohort. However, the period effect is only strongly evident in Whites. Visual inspection of White trends suggests possible differences for those older and younger than age 50. These methods are applied in an interactive website displaying incidence and mortality trends for over 20 cancer sites in the US.

Toward quantitative CEST imaging of glutamate in the mouse brain using a multi-pool exchange model calibrated by 1H-MRS.

To develop a CEST quantification model to map glutamate concentration in the mouse brain at 11.7 T, overcoming the limitations of conventional glutamate-weighted CEST (gluCEST) contrast (magnetization transfer ratio with asymmetric analysis). 1H-MRS was used as a gold standard for glutamate quantification to calibrate a CEST-based quantitative pipeline. Joint localized measurements of Z-spectra at B1 = 5 μT and quantitative 1H-MRS were carried out in two voxels of interest in the mouse brain. A six-pool Bloch-McConnell model was found appropriate to fit experimental data. Glutamate exchange rate was estimated in both regions with this dedicated multi-pool fitting model and using glutamate concentration determined by 1H-MRS. Glutamate exchange rate was estimated to be ˜1300 Hz in the mouse brain. Using this calibrated value, maps of glutamate concentration in the mouse brain were obtained by pixel-by-pixel fitting of Z-spectra at B1 = 5 μT. A complementary study of simulations, however, showed that the quantitative model has high sensitivity to noise, and therefore, requires high-SNR acquisitions. Interestingly, fitted [Glu] seemed to be overestimated compared to 1H-MRS measurements, although it was estimated with simulations that the model has no intrinsic fitting bias with our experimental level of noise. The hypothesis of an unknown proton-exchanging pool contributing to gluCEST signal is discussed. High-resolution mapping of glutamate in the brain was made possible using the proposed calibrated quantification model of gluCEST data. Further studying of the in vivo molecular contributions to gluCEST signal could improve modeling.

Combining Magnetostriction with Variable Reluctance for Energy Harvesting at Low Frequency Vibrations

In this paper, we explore the benefits of using a magnetostrictive component in a variable reluctance energy harvester. The intrinsic magnetic field bias and the possibility to utilize magnetic force to achieve pre-stress leads to a synergetic combination between this type of energy harvester and magnetostriction. The proposed energy harvester system, to evaluate the concept, consists of a magnetostrictive cantilever beam with a cubic magnet as proof mass. Galfenol, Fe81.6Ga18.4, is used to implement magnetostriction. Variable reluctance is achieved by fixing the beam parallel to an iron core, with some margin to create an air gap between the tip magnet and core. The mechanical forces of the beam and the magnetic forces lead to a displaced equilibrium position of the beam and thus a pre-stress. Two configurations of the energy harvester were evaluated and compared. The initial configuration uses a simple beam of aluminum substrate and a layer of galfenol with an additional magnet fixing the beam to the core. The modified design reduces the magnetic field bias in the galfenol by replacing approximately half of the length of galfenol with aluminum and adds a layer of soft magnetic material above the galfenol to further reduce the magnetic field bias. The initial system was found to magnetically saturate the galfenol at equilibrium. This provided the opportunity to compare two equivalent systems, with and without a significant magnetostrictive effect on the output voltage. The resonance frequency tuning capability, from modifying the initial distance of the air gap, is shown to be maintained for the modified configuration (140 Hz/mm), while achieving RMS open-circuit coil voltages larger by a factor of two (2.4 V compared to 1.1 V). For a theoretically optimal load, the RMS power was simulated to be 5.1 mW. Given the size of the energy harvester (18.5 cm3) and the excitation acceleration (0.5 g), this results in a performance metric of 1.1 mW/cm3g2.

Robust unsupervised 5D seismic data reconstruction on regular and irregular grids

Seismic data reconstruction has become a central focus in seismic data processing, addressing challenges posed by sparse sampling due to physical and budgetary constraints. The advent of 5D acquisition methodologies marks a significant advancement in the quality and completeness of seismic data sets. Most traditional 5D reconstruction methods commonly use the fast Fourier transform (FFT), requiring regular grids and preliminary 4D binning before 5D interpolation. Discrete Fourier transform and nonequidistant FFT can honor the original irregular coordinates. However, when using exact locations, these methods become computationally expensive. We introduce an unsupervised deep-learning methodology to learn a continuous function across the sampling points in seismic data, facilitating reconstruction on regular and irregular grids. The network comprises a multilayer perceptron with linear layers and element-wise periodic activation functions. It excels at mapping the input coordinates to the corresponding seismic data amplitudes without relying on external training sets. The network’s intrinsic low-frequency bias is crucial in prioritizing acquiring self-similar features over high-frequency and incoherent ones during training. This characteristic mitigates incoherent noise in seismic data, such as random and erratic components. To assess the robustness of the unsupervised reconstruction technique, we conduct comprehensive evaluations using synthetic data examples sampled regularly and irregularly, as well as field-data examples with and without binning. The findings demonstrate the efficacy of our deep-learning framework in achieving resilient and accurate seismic data reconstruction across diverse sampling scenarios.

Neural crest precursors from the skin are the primary source of directly reprogrammed neurons

Direct reprogramming involves the conversion of differentiated cell types without returning to an earlier developmental state. Here, we explore how heterogeneity in developmental lineage and maturity of the starting cell population contributes to direct reprogramming using the conversion of murine fibroblasts into neurons. Our hypothesis is that a single lineage of cells contributes to most reprogramming and that a rare elite precursor with intrinsic bias is the source of reprogrammed neurons. We find that nearly all reprogrammed neurons are derived from the neural crest (NC) lineage. Moreover, when rare proliferating NC precursors are selectively ablated, there is a large reduction in the number of reprogrammed neurons. Previous interpretations of this paradigm are that it demonstrates a cell fate conversion across embryonic germ layers (mesoderm to ectoderm). Our interpretation is that this is actually directed differentiation of a neural lineage stem cell in the skin that has intrinsic bias to produce neuronal progeny.

A Massively Parallel In Vivo Assay of TdT Mutants Yields Variants with Altered Nucleotide Insertion Biases.

Terminal deoxynucleotidyl transferase (TdT) is a unique DNA polymerase capable of template-independent extension of DNA. TdT's de novo DNA synthesis ability has found utility in DNA recording, DNA data storage, oligonucleotide synthesis, and nucleic acid labeling, but TdT's intrinsic nucleotide biases limit its versatility in such applications. Here, we describe a multiplexed assay for profiling and engineering the bias and overall activity of TdT variants with high throughput. In our assay, a library of TdTs is encoded next to a CRISPR-Cas9 target site in HEK293T cells. Upon transfection of Cas9 and sgRNA, the target site is cut, allowing TdT to intercept the double-strand break and add nucleotides. Each resulting insertion is sequenced alongside the identity of the TdT variant that generated it. Using this assay, 25,623 unique TdT variants, constructed by site-saturation mutagenesis at strategic positions, were profiled. This resulted in the isolation of several altered-bias TdTs that expanded the capabilities of our TdT-based DNA recording system, Cell HistorY Recording by Ordered InsertioN (CHYRON), by increasing the information density of recording through an unbiased TdT and achieving dual-channel recording of two distinct inducers (hypoxia and Wnt) through two differently biased TdTs. Select TdT variants were also tested in vitro, revealing concordance between each variant's in vitro bias and the in vivo bias determined from the multiplexed high throughput assay. Overall, our work and the multiplex assay it features should support the continued development of TdT-based DNA recorders, in vitro applications of TdT, and further study of the biology of TdT.

Intrinsic exchange bias from interfacial reconstruction in an epitaxial NixCoyFe3−x−yO4(111)/α-Al2O3(0001) thin film family

Intrinsic exchange bias is known as the unidirectional exchange anisotropy that emerges in a nominally single-component ferro-(ferri-)magnetic system. In this work, with magnetic and structural characterizations, we demonstrate that intrinsic exchange bias is a general phenomenon in (Ni, Co, Fe)-based spinel oxide films deposited onα-Al2O3(0001) substrates, due to the emergence of a rock-salt interfacial layer consisting of antiferromagnetic CoO from interfacial reconstruction. We show that in NixCoyFe3-x-yO4(111)/α-Al2O3(0001) films, intrinsic exchange bias and interfacial reconstruction have consistent dependences on Co concentrationy, while the Ni and Fe concentration appears to be less important. This work establishes a family of intrinsic exchange bias materials with great tunability by stoichiometry and highlights the strategy of interface engineering in controlling material functionalities.

Prebiotic chiral transfer from self-aminoacylating ribozymes may favor either handedness

Modern life is essentially homochiral, containing D-sugars in nucleic acid backbones and L-amino acids in proteins. Since coded proteins are theorized to have developed from a prebiotic RNA World, the homochirality of L-amino acids observed in all known life presumably resulted from chiral transfer from a homochiral D-RNA World. This transfer would have been mediated by aminoacyl-RNAs defining the genetic code. Previous work on aminoacyl transfer using tRNA mimics has suggested that aminoacylation using D-RNA may be inherently biased toward reactivity with L-amino acids, implying a deterministic path from a D-RNA World to L-proteins. Using a model system of self-aminoacylating D-ribozymes and epimerizable activated amino acid analogs, we test the chiral selectivity of 15 ribozymes derived from an exhaustive search of sequence space. All of the ribozymes exhibit detectable selectivity, and a substantial fraction react preferentially to produce the D-enantiomer of the product. Furthermore, chiral preference is conserved within sequence families. These results are consistent with the transfer of chiral information from RNA to proteins but do not support an intrinsic bias of D-RNA for L-amino acids. Different aminoacylation structures result in different directions of chiral selectivity, such that L-proteins need not emerge from a D-RNA World.

A novel multi-modal fusion method based on uncertainty-guided meta-learning

Multi-modal data fusion for effective feature representation in machine learning is challenging due to intrinsic biases present within and across different modalities. Existing multi-modal data fusion methods often face difficulties in learning generic features due to diverse noise patterns and variations in feature dynamics across different modalities. In this paper, we present a novel method called Uncertainty-guided Meta-Learning Multi-modal Fusion and Classification (UMLMC) to address these challenges. UMLMC dynamically transforms multi-modal feature spaces at both the pre- and post-fusion levels by incorporating uncertainty estimates from an auxiliary network. Our model is optimized using a meta-learning algorithm to enhance its generalization capabilities. Extensive experiments on multi-modal data from diverse domains, along with comparisons to state-of-the-art methods, demonstrate the effectiveness of UMLMC in improving classification performance. These results confirm that UMLMC, with its innovative uncertainty estimation and meta-learning framework, effectively learns informative intra- and inter-modal features, leading to superior classification outcomes.

An integrated transcription factor framework for Treg identity and diversity.

Vertebrate cell identity depends on the combined activity of scores of transcription factors (TF). While TFs have often been studied in isolation, a systematic perspective on their integration has been missing. Focusing on FoxP3+ regulatory T cells (Tregs), key guardians of immune tolerance, we combined single-cell chromatin accessibility, machine learning, and high-density genetic variation, to resolve a validated framework of diverse Treg chromatin programs, each shaped by multi-TF inputs. This framework identified previously unrecognized Treg controllers (Smarcc1) and illuminated the mechanism of action of FoxP3, which amplified a pre-existing Treg identity, diversely activating or repressing distinct programs, dependent on different regulatory partners. Treg subpopulations in the colon relied variably on FoxP3, Helios+ Tregs being completely dependent, but RORγ+ Tregs largely independent. These differences were rooted in intrinsic biases decoded by the integrated framework. Moving beyond master regulators, this work unravels how overlapping TF activities coalesce into Treg identity and diversity.

On the Correlation between Young Massive Star Clusters and Gamma-Ray Unassociated Sources

Star clusters (SCs) are potential cosmic-ray accelerators and therefore are expected to emit high-energy radiation. However, a clear detection of gamma-ray emission from this source class has only been possible for a handful of cases. This could in principle result from two different reasons: either detectable SCs are limited to a small fraction of the total number of Galactic SCs, or gamma-ray-emitting SCs are not recognized as such and therefore are listed in the ensemble of unidentified sources. In this Letter we investigate this latter scenario by comparing available catalogs of SCs and H ii regions, obtained from Gaia and Wide-field Infrared Survey Explorer observations, to the gamma-ray GeV and TeV catalogs built from Fermi Large Area Telescope (LAT), H.E.S.S., and LHAASO data. The significance of the correlation between catalogs is evaluated by comparing the results with simulations of synthetic populations. A strong correlation emerges between Fermi-LAT-unidentified sources and H ii regions that trace massive SCs in the earliest (≲1–2 Myr) phase of their life, where no supernova explosions have happened yet, confirming that winds of massive stars can alone accelerate particles and produce gamma-ray emission at least up to GeV energies. The association with TeV energy sources is less evident. Similarly, no significant association is found between Gaia SCs and GeV nor TeV sources. We ascribe this fact to the larger extension of these objects but also to an intrinsic bias in the Gaia selection toward SCs surrounded by a lower target gas density, which would otherwise hinder the detection in the optical wave band.

Demystifying and Mitigating Bias for Node Representation Learning.

Node representation learning has attracted increasing attention due to its efficacy for various applications on graphs. However, fairness is a largely under-explored territory within the field, although it is shown that the use of graph structure in learning amplifies bias. To this end, this work theoretically explains the sources of bias in node representations obtained via graph neural networks (GNNs). It is revealed that both nodal features and graph structure lead to bias in the obtained representations. Building upon the analysis, fairness-aware data augmentation frameworks are developed to reduce the intrinsic bias. Our theoretical analysis and proposed schemes can be readily employed in understanding and mitigating bias for various GNN-based learning mechanisms. Extensive experiments on node classification and link prediction over multiple real networks are carried out, and it is shown that the proposed augmentation strategies can improve fairness while providing comparable utility to state-of-the-art methods.

A multifaceted approach to detect gender biases in Natural Language Generation

Recent advances in generative models have skyrocketed the popularity of conversational chatbots and have revolutionized the way we interact with artificial intelligence. At the same time, research has shown that machine learning models can unconsciously reflect and amplify human biases. This is particularly dangerous for generative models given the huge popularity of such technologies. Specifically, a fundamental source of bias of such technologies is the resources on which the models are trained. To address this issue, this paper proposes a methodology to analyze intrinsic gender bias in Natural Language Generation (NLG). Some works already propose metrics and approaches to measure bias in the Natural Language processing field. However, there is a lack of standard methodology to measure gender bias in NLG. Therefore, adapting the Bias Score approach, our proposal involves three sequential stages applied to individual texts to detect intrinsic gender bias on NLG effectively. Those steps are as follows: (i) word scoring; (ii) word filtering; and (iii) generative-word analysis. This methodology is applied to recent datasets and pre-trained models widely used for the generation of text with common sense. In particular, this paper analyzes the potential gender bias in CommonGen and C2Gen datasets and the SimpleNLG and T5 models. The results show the ability of the proposed methodology to detect gender bias in word distributions, presenting a strong correlation with the words typically associated with a specific gender. Results indicate that both tested datasets are intrinsically gender-biased, and therefore, tested models fine-tuned with those datasets also are.

The sugar cube: Network control and emergence in stereoediting reactions.

Stereochemical editing strategies have recently enabled the transformation of readily accessible substrates into rare and valuable products. Typically, site selectivity is achieved by minimizing kinetic complexity by using protecting groups to suppress reactivity at undesired sites (substrate control) or by using catalysts with tailored shapes to drive reactivity at the desired site (catalyst control). We propose "network control," a contrasting paradigm that exploits hidden interactions between rate constants to greatly amplify modest intrinsic biases and enable precise multisite editing. When network control is applied to the photochemical isomerization of hexoses, six of the eight possible diastereomers can be selectively obtained. The amplification effect can be viewed as a mesoscale phenomenon between the limiting regimes of kinetic control in simple chemical systems and metabolic regulation in complex biological systems.

How to reward animals based on their subjective percepts: A Bayesian approach to online estimation of perceptual biases.

Elucidating the neural basis of perceptual biases, such as those produced by visual illusions, can provide powerful insights into the neural mechanisms of perceptual inference. However, studying the subjective percepts of animals poses a fundamental challenge: unlike human participants, animals cannot be verbally instructed to report what they see, hear, or feel. Instead, they must be trained to perform a task for reward, and researchers must infer from their responses what the animal perceived. However, animals' responses are shaped by reward feedback, thus raising the major concern that the reward regimen may alter the animal's decision strategy or even intrinsic perceptual biases. We developed a method that estimates perceptual bias during task performance and then computes the reward for each trial based on the evolving estimate of the animal's perceptual bias. Our approach makes use of multiple stimulus contexts to dissociate perceptual biases from decision-related biases. Starting with an informative prior, our Bayesian method updates a posterior over the perceptual bias after each trial. The prior can be specified based on data from past sessions, thus reducing the variability of the online estimates and allowing it to converge to a stable estimate over a small number of trials. After validating our method on synthetic data, we apply it to estimate perceptual biases of monkeys in a motion direction discrimination task in which varying background optic flow induces robust perceptual biases. This method overcomes an important challenge to understanding the neural basis of subjective percepts.

An allocentric human odometer for perceiving distances on the ground plane.

We reliably judge locations of static objects when we walk despite the retinal images of these objects moving with every step we take. Here, we showed our brains solve this optical illusion by adopting an allocentric spatial reference frame. We measured perceived target location after the observer walked a short distance from the home base. Supporting the allocentric coding scheme, we found the intrinsic bias , which acts as a spatial reference frame for perceiving location of a dimly lit target in the dark, remained grounded at the home base rather than traveled along with the observer. The path-integration mechanism responsible for this can utilize both active and passive (vestibular) translational motion signals, but only along the horizontal direction. This asymmetric path-integration finding in human visual space perception is reminiscent of the asymmetric spatial memory finding in desert ants, pointing to nature's wondrous and logically simple design for terrestrial creatures.

Exploring collaborative decision-making: A quasi-experimental study of human and Generative AI interaction

This paper explores the effects of integrating Generative Artificial Intelligence (GAI) into decision-making processes within organizations, employing a quasi-experimental pretest-posttest design. The study examines the synergistic interaction between Human Intelligence (HI) and GAI across four group decision-making scenarios within three global organizations renowned for their cutting-edge operational techniques. The research progresses through several phases: identifying research problems, collecting baseline data on decision-making, implementing AI interventions, and evaluating the outcomes post-intervention to identify shifts in performance. The results demonstrate that GAI effectively reduces human cognitive burdens and mitigates heuristic biases by offering data-driven support and predictive analytics, grounded in System 2 reasoning. This is particularly valuable in complex situations characterized by unfamiliarity and information overload, where intuitive, System 1 thinking is less effective. However, the study also uncovers challenges related to GAI integration, such as potential over-reliance on technology, intrinsic biases particularly ‘out-of-the-box’ thinking without contextual creativity. To address these issues, this paper proposes an innovative strategic framework for HI-GAI collaboration that emphasizes transparency, accountability, and inclusiveness.