Representativeness Research Articles

Simple Debiased Contrastive Learning for Sequential Recommendation

Recently, contrastive learning has been shown to be effective in improving sequential recommendation, alleviating noise and long-tail issues caused by real-world noise and data sparsity, and enabling models to learn high-quality user representations. The data augmentation approach of contrastive learning mainly includes randomly changing the sequence data and adding randomly perturbations, and by pulling the distance between the original sample and the positive sample and pushing the distance between the original sample and the negative sample to enhance alignment and to optimize the uniformity of representation space. However, these approaches may use false positive/negative samples, learn biased user representations, and compromise the uniformity of the representation space, leading to second-best recommendation results. To solve these issues, we propose a generic Simple Debiased Contrastive Learning for Sequential Recommendation framework (SimDCL), to alleviate these biased samples, optimize the uniformity of the representation space, and enhance anti-noise ability. Specifically, we augment the samples based on noise, and optimize the uniformity of the representation space through gradient algorithms, and design an object filtering approach to punish false positive and false negative samples in user interaction sequences. Finally, we combine contrastive learning and sequential recommendation for multi-task collaborative training to improve the model’s anti-noise capability and the quality of recommendation. The empirical studies on four benchmark datasets prove the superiority of our approach, and the code and datasets are available at URL.

Gene representation bias in spatial transcriptomics.

For sequencing-based spatial transcriptomics data, the gene-spot count matrix is highly sparse. This feature is similar to scRNA-seq. The goal of this paper is to identify whether there exist genes that are frequently under-detected in Visium compared to bulk RNA-seq, and the underlying potential mechanism of under-detection in Visium. We collected paired Visium and bulk RNA-seq data for 28 human samples and 19 mouse samples, which covered diverse tissue sources. We compared the two data types and observed that there indeed exists a collection of genes frequently under-detected in Visium compared to bulk RNA-seq. We performed a motif search to examine the last 350 bp of the frequently under-detected genes, and we observed that the poly (T) motif was significantly enriched in genes identified from both human and mouse data, which matches with our previous finding about frequently under-detected genes in scRNA-seq. We hypothesized that the poly (T) motif may be able to form a hairpin structure with the poly (A) tails of their mRNA transcripts, making it difficult for their mRNA transcripts to be captured during Visium library preparation.

Evolving and Novel Applications of Artificial Intelligence in Thoracic Imaging.

The advent of artificial intelligence (AI) is revolutionizing medicine, particularly radiology. With the development of newer models, AI applications are demonstrating improved performance and versatile utility in the clinical setting. Thoracic imaging is an area of profound interest, given the prevalence of chest imaging and the significant health implications of thoracic diseases. This review aims to highlight the promising applications of AI within thoracic imaging. It examines the role of AI, including its contributions to improving diagnostic evaluation and interpretation, enhancing workflow, and aiding in invasive procedures. Next, it further highlights the current challenges and limitations faced by AI, such as the necessity of 'big data', ethical and legal considerations, and bias in representation. Lastly, it explores the potential directions for the application of AI in thoracic radiology.

Insufficient and biased representation of species geographic responses to climate change.

The geographic redistributions of species due to a rapidly changing climate are poised to perturb ecological communities and significantly impact ecosystems and human livelihoods. Effectively managing these biological impacts requires a thorough understanding of the patterns and processes of species geographic range shifts. While substantial recent redistributions have been identified and recognized to vary by taxon, region, and range geometry, there are large gaps and biases in the available evidence. Here, we use the largest compilation of geographic range change observations to date, comprised of 33,016 potential redistributions across 12,009 species, to formally assess within- and cross-species coverage and biases and to motivate future data collection. We find that species coverage varies strongly by taxon and underrepresents species at high and low latitudes. Within species, assessments of potential redistributions came from parts of their geographic range that were highly uneven and non-representative. For most species and taxa, studies were strongly biased toward the colder parts of species' distributions and thus significantly underrepresented populations that might get pushed beyond their maximum temperature limits. Coverage of potential leading and trailing geographic range edges under a changing climate was similarly uneven. Only 8% of studied species were assessed at both high and low latitude and elevation range edges, with most only covered at one edge. This suggests that substantial within-species biases exacerbate the considerable geographic and taxonomic among-species unevenness in evidence. Our results open the door for a more quantitative accounting for existing knowledge biases in climate change ecology and a more informed management and conservation. Our findings offer guidance for future data collection that better addresses information gaps and provides a more effective foundation for managing the biological impacts of climate change.

Global Threads, Unveiling Unevenness: Contemporary Maximalist Projects Interrogating Cultural Hybridisation and Marginality

Global Threads, Unveiling Unevenness: Contemporary Maximalist Projects Interrogating Cultural Hybridisation and Marginality. Within a frame that emphasizes the tension between the global and the local, this paper aims to investigate the ways in which complex narratives that incorporate the tropes of migration, periphery, and marginality, amongst others, can bring to light aspects of unevenness and cultural and formal hybridisation. Works like Zadie Smith’s White Teeth, Arundhati Roy’s The Ministry of Utmost Happiness, or Bernardine Evaristo’s Girl, Woman, Other employ distinct maximalist modes of inquiry (Nick Levey) to question topics related to inequality, cultural relevance, or representational biases, in a type of novel about which James Wood claimed that it “suffered from an excess of storytelling and an almost paranoid preoccupation with linking up their many subplots in a web of forced meaning.” What stands at the core of this article is precisely this impulse to force a meaning which seems most frequently disrupted by an anarchetypal propensity to renegotiate a “rhetoric of inclusivity” (Franco Moretti) through which the maximalist author tries to exhaustively encompass the whole world (Levey). By selecting a corpus of maximalist novels to illustrate their evolution from the second half of the 20th century until more recent works, such as Evaristo’s, this paper investigates the shift through which these narratives have started to factor margins in, differently and more frequently than in the beginnings of this literary form. Keywords: Maximalist Novels; Hybridisation; Migration; Periphery; Marginality; Anarchetype; Postmodernism.

North American cooling signature of strong stratospheric wave events depends on the QBO phase

Extreme stratospheric wave activity has been linked to surface cold extremes over North America, but little is known whether the Quasi-biennial Oscillation (QBO) plays a role in this linkage. Here, by comparing strong stratospheric wave events during the westerly phase (wQBO) with those during the easterly phase (eQBO), we show that the cooling signature following strong wave events depends on the QBO phase in observations. During wQBO, strong wave events are followed by an increased risk of North American cold extremes and a vertical structure shift from a westward phase tilt to an eastward tilt. However, strong wave events under eQBO do not change the cold risk nor alter the vertical tilt. We further examine this dependence on QBO in QBO-resolving climate models, finding that the cooling signature of strong wave events in models is largely insensitive to QBO phases. This insensitivity is suggested to be linked to model biases in the stratospheric wave representation.

Explaining Neural News Recommendation with Attributions onto Reading Histories

An important aspect of responsible recommendation systems is the transparency of the prediction mechanisms. This is a general challenge for deep-learning-based systems such as the currently predominant neural news recommender architectures which are optimized to predict clicks by matching candidate news items against users’ reading histories. Such systems achieve state-of-the-art click-prediction performance, but the rationale for their decisions is difficult to assess. At the same time, the economic and societal impact of these systems makes such insights very much desirable. In this paper, we ask the question to what extent the recommendations of current news recommender systems are actually based on content-related evidence from reading histories. We approach this question from an explainability perspective. Building on the concept of integrated gradients, we present a neural news recommender that can accurately attribute individual recommendations to news items and words in input reading histories while maintaining a top scoring click-prediction performance. Using our method as a diagnostic tool, we find that: (a), a substantial number of users’ clicks on news are not explainable from reading histories, and many history-explainable items are actually skipped; (b), while many recommendations are based on content-related evidence in histories, for others the model does not attend to reasonable evidence, and recommendations stem from a spurious bias in user representations. Our code is publicly available 1 .

Deep Fusion Clustering Network With Reliable Structure Preservation.

Deep clustering, which can elegantly exploit data representation to seek a partition of the samples, has attracted intensive attention. Recently, combining auto-encoder (AE) with graph neural networks (GNNs) has accomplished excellent performance by introducing structural information implied among data in clustering tasks. However, we observe that there are some limitations of most existing works: 1) in practical graph datasets, there exist some noisy or inaccurate connections among nodes, which would confuse network learning and cause biased representations, thus leading to unsatisfied clustering performance; 2) lacking dynamic information fusion module to carefully combine and refine the node attributes and the graph structural information to learn more consistent representations; and 3) failing to exploit the two separated views' information for generating a more robust target distribution. To solve these problems, we propose a novel method termed deep fusion clustering network with reliable structure preservation (DFCN-RSP). Specifically, the random walk mechanism is introduced to boost the reliability of the original graph structure by measuring localized structure similarities among nodes. It can simultaneously filter out noisy connections and supplement reliable connections in the original graph. Moreover, we provide a transformer-based graph auto-encoder (TGAE) that can use a self-attention mechanism with the localized structure similarity information to fine-tune the fused topology structure among nodes layer by layer. Furthermore, we provide a dynamic cross-modality fusion strategy to combine the representations learned from both TGAE and AE. Also, we design a triplet self-supervision strategy and a target distribution generation measure to explore the cross-modality information. The experimental results on five public benchmark datasets reflect that DFCN-RSP is more competitive than the state-of-the-art deep clustering algorithms. The corresponding code is available at https://github.com/gongleii/DFCN-RSP.

Deep-GenMut: Automated genetic mutation classification in oncology: A deep learning comparative study

Early cancer detection and treatment depend on the discovery of specific genes that cause cancer. The classification of genetic mutations was initially done manually. However, this process relies on pathologists and can be a time-consuming task. Therefore, to improve the precision of clinical interpretation, researchers have developed computational algorithms that leverage next-generation sequencing technologies for automated mutation analysis. This paper utilized four deep learning classification models with training collections of biomedical texts. These models comprise bidirectional encoder representations from transformers for Biomedical text mining (BioBERT), a specialized language model implemented for biological contexts. Impressive results in multiple tasks, including text classification, language inference, and question answering, can be obtained by simply adding an extra layer to the BioBERT model. Moreover, bidirectional encoder representations from transformers (BERT), long short-term memory (LSTM), and bidirectional LSTM (BiLSTM) have been leveraged to produce very good results in categorizing genetic mutations based on textual evidence. The dataset used in the work was created by Memorial Sloan Kettering Cancer Center (MSKCC), which contains several mutations. Furthermore, this dataset poses a major classification challenge in the Kaggle research prediction competitions. In carrying out the work, three challenges were identified: enormous text length, biased representation of the data, and repeated data instances. Based on the commonly used evaluation metrics, the experimental results show that the BioBERT model outperforms other models with an F1 score of 0.87 and 0.850 MCC, which can be considered as improved performance compared to similar results in the literature that have an F1 score of 0.70 achieved with the BERT model.

Overcoming Fractures of Cross-Cultural Understanding: Modern Chinese Knowledge Production on the Arab World

Despite the long history of Chinese-Arab interactions since the seventh century, modern Chinese scholarship on the Arab world has a much shorter story. The global historical context that led to such a situation is the Western dominance in knowledge production. Most Chinese scholars research on issues related to the West and/or China-Western interactions. Only a small proportion of researchers cast their eyes on the “other East”—the Arab world. Consequently, Arab/Arabic studies occupies a marginal space in Chinese academies. In addition, although Edward Said had fiercely argued more than forty years ago in Orientalism that subtle but persistent Eurocentric prejudice against Arab-Islamic peoples and their culture perpetrates Western misunderstanding about the region, such biased representations were sometimes uncritically transferred by Chinese scholars when they were acquiring knowledge about the region via translations of Orientalist accounts written in English. Another major obstacle is the inherent linguistic and orthographic difficulties involved. Although some researchers have made great endeavors to overcome such challenges, most publish only in Chinese. As a result, their findings are not widely known. This article, therefore, scrutinizes and contextualizes modern Chinese scholarship on the Arab world in order to introduce it to a wider international audience, especially for those who are keenly observing the increasing Chinese-Arab engagements in the twenty-first century. In addition to providing synoptic overviews of major institutions, scholars, and their representative works, the article also critically analyzes the historical contexts that led to the initial formation, subsequent divergent developments in Chinese-language academies on both sides of the Taiwan Strait, and the present challenges of the respective state of the fields. I argue that due to bigger population, larger geographical size, and more government funding, Arab/Arabic studies in mainland China is more developed than that in Taiwan, though scholars based on the mainland are facing more pressure on censorship and self-censorship. Both fields, however, were pioneered by the same group of Chinese Hui Muslim scholars who went to study at al-Azhar University in Egypt in the 1930s and the 1940s, not unlike cosmopolitan Muslims around the world at the time who traveled to Cairo for multiple and complex reasons. When they returned, institutionalizing Arab/Arabic-Islamic studies in Chinese academies became one of their most visible and long-lasting legacies. Contingencies in their lives, mainly job opportunities upon graduation, played an important role in their later bifurcated career trajectories. The process during which these Hui Azharites ceased to be “cosmopolitan Muslims” and instead became “Chinese Arabists” reflects an important transformation in Chinese-Arab interactions during the Cold War politics. Later generations of scholars developed systematic pedagogies on teaching Arabic to native Chinese speakers by publishing Arabic-Chinese dictionaries as well as textbooks and grammar books. Their role as cross-cultural intermediaries is significant and widely impactful. These little-known scholars deserve to be recognized for their intellectual pursuits.

The Effects of Equiprobability Bias and Representativeness Heuristics on the Performance in Probability Comparison and Calculation Tasks Among Middle School Students in China

The Effects of Equiprobability Bias and Representativeness Heuristics on the Performance in Probability Comparison and Calculation Tasks Among Middle School Students in China

A review of explainable fashion compatibility modeling methods

The paper reviews methods used in the fashion compatibility recommendation domain. We select methods based on reproducibility, explainability, and novelty aspects and then organize them chronologically and thematically. We presented general characteristics of publicly available datasets that are related to the fashion compatibility recommendation task. Finally, we analyzed the representation bias of datasets, fashion-based algorithms’ sustainability, and explainable model assessment. The paper describes practical problem explanations, methodologies, and published datasets that may serve as an inspiration for further research. The proposed structure of the survey organizes knowledge in the fashion recommendation domain and will be beneficial for those who want to learn the topic from scratch, expand their knowledge, or find a new field for research. Furthermore, the information included in this paper could contribute to developing an effective and ethical fashion-based recommendation system.

Investigation of model forecast biases and skilful prediction for Assam heavy rainfall 2022

Extreme rainfall events (ERE) during the summer monsoon season have been occurring over most parts of India resulting in flooding and immense socio-economic loss. These extremes are becoming a frequent norm in the hilly and mountainous regions of the country such as Assam. Assam received one of the most historical EREs from 14–June 17, 2022. The present study analyses the performance of a suite of high-resolution ensemble model forecasts for this extreme event in terms of its intensity, and distribution with a lead time of up to 96 h. Furthermore, the 36 numerical experiments are carried out using two different land use and land cover (LULC) data sets (i.e. ISRO and USGS) and three different sets of parameterization schemes (i.e. planetary boundary layer, cumulus, and microphysics).Rainfall distributions in the case of USGS LULC are relatively less coherent and underestimated (60–260 mm/day) against IMD (80–300 mm/day) including the rainfall categories heavy (HR), very heavy (VHR), and extremely heavy (EHR) rainfall throughout the day-1 to day-4. Among all the ensembles (E1-E10), USGS (E6 - E10) has underestimated rainfall (140–260 mm/day) compared to ISRO (150–280 mm/day), specifically in MR and HR categories over the upper Assam (UAD) and lower Assam (LAD) divisions. Further, the Bias Correction Ensemble (BCE) technique is applied to minimize the forecast errors. A rigorous statistical analysis in terms of frequency distribution, Taylor diagram, and benchmark skill scores is carried out to elucidate the model biases. The set of the model ensembles using ISRO (E1- E5) and USGS (E6- E10) reasonably captured the HR, VHR, and EHR. In addition, throughout the forecast hour, BCE E5 (E10) is noted with the distinct realistic (underestimated) representation of model bias (5–20 %) (10–30 %) over all the subdivisions of Assam. Our results suggest that the combined efforts of ensembles of physical parameterization schemes, along with proper LULC, and the BCE approach are required to overcome challenges to improve the skills of rainfall events, particularly over complex terrains such as Assam.

Memory retrieval effects as a function of differences in phenomenal experience.

Conscious experience and perception are restricted to a single perspective. Although evidence to suggest differences in phenomenal experience can produce observable differences in behavior, it is not well understood how these differences might influence memory. We used fMRI to scan n = 49 participants while they encoded and performed a recognition memory test for faces and words. We calculated a cognitive bias score reflecting individual participants' propensity toward either Visual Imagery or Internal Verbalization based on their responses to the Internal Representations Questionnaire (IRQ). Neither visual imagery nor internal verbalization scores were significantly correlated with memory performance. In the fMRI data, there were typical patterns of activation differences between words and faces during both encoding and retrieval. There was no effect of internal representation bias on fMRI activation during encoding. At retrieval, however, a bias toward visualization was positively correlated with memory-related activation for both words and faces in inferior occipital gyri. Further, there was a crossover interaction in a network of brain regions such that visualization bias was associated with greater activation for words and verbalization bias was associated with greater activation for faces, consistent with increased effort for non-preferred stimulus retrieval. These findings suggest that individual differences in cognitive representations affect neural activation across different types of stimuli, potentially affecting memory retrieval performance.

Perpetuation of Gender Bias in Visual Representation of Professions in the Generative AI Tools DALL·E and Bing Image Creator

Artificial intelligence (AI)-based generative imaging systems such as DALL·E, Midjourney, Stable Diffusion, and Adobe Firefly, which work by transforming natural language descriptions into images, are revolutionizing computer vision. In this exploratory and qualitative research, we have replicated requests for images of women in different professions by comparing these representations in previous studies with DALL·E, observing that this model continues to provide in its last version, DALL·E 3, inequitable results in terms of gender. In addition, Bing Image Creator, Microsoft’s free tool that is widely used among the population and runs under DALL·E, has been tested for the first time. It also presents a sexualization of women and stereotypical children’s representations. The results reveal the following: 1. A slight improvement in terms of the presence of women in professions previously shown only with men. 2. They continue to offer biased results in terms of the objectification of women by showing sexualized women. 3. The representation of children highlights another level of gender bias, reinforcing traditional stereotypes associated with gender roles from childhood, which can impact future decisions regarding studies and occupations.

Causes of Past African Temperature Change in PMIP Simulations of the Mid‐Holocene

AbstractCurrent‐generation climate models project that Africa will warm by up to 5°C in the coming century, severely stressing African populations. Past and ongoing work indicates, however, that the models used to create these projections do not match proxy records of past temperature in Africa during the mid‐Holocene (MH), raising concerns that their future projections may house large uncertainties. Rather than reproducing proxy‐based reconstructions of MH warming relative to the Pre‐Industrial (PI), models instead simulate MH temperatures very similar to or slightly colder than the PI. This data‐model mismatch could be due to a variety of factors, including biases in model surface energy budgets or inaccurate representation of the feedbacks between temperature and hydrologic change during the “Green Sahara.” We focus on the differences among model simulations in the Paleoclimate Modeling Intercomparison Project Phases 3 and 4 (PMIP3 and PMIP4), examining surface temperature and energy budgets to investigate controls on temperature and the potential model sources of this paleoclimate data‐model mismatch. Our results suggest that colder conditions simulated by PMIP3 and PMIP4 models during the MH are in large part due to the joint impacts of feedback uncertainties in response to increased precipitation, a strengthened West African Monsoon (WAM) in the Sahel, and the Green Sahara. We extend these insights into suggestions for model physics and boundary condition changes, and discuss implications for the accuracy of future climate model projections over Africa.

Mitigating biases in long-tailed recognition via semantic-guided feature transfer

Dealing with significant class imbalance poses a significant challenge in various real-world applications, particularly when the accurate classification and generalization of minority classes are of crucial. One key factor that results in model biases is the inadequate representation of intra-class diversity in samples from tail classes. To tackle this challenge, transfer learning strategies have emerged as a viable approach to promote the data distribution of tail classes by transferring feature learned from head classes. However, existing methods often assume that transferability exists between any head class and tail class, and randomly transfer information during training, which can be problematic when the head and tail classes have uncorrelated semantics. To tackle this problem, we propose a novel feature transfer strategy to promote the learning of discriminative representations for tail classes. Firstly, we compute the semantic similarity in a hybrid manner by combining local visual features and global semantic features. Secondly, we utilize the Gaussian mixture model to represent each class with multiple and anisotropic prototypes, which helps us fit the actual distribution accurately. We then carry out distribution-aware head-to-tail feature transfer, leveraging an optimization objective that can derive a computationally efficient closed-form upper bound. This procedure mitigates representation biases in the feature space and achieves state-of-the-art results on CIFAR100-LT, ImageNet-LT, and iNaturalist long-tailed benchmarks.

Bias correction in species distribution models based on geographic and environmental characteristics

Correcting sampling bias in species distribution models (SDMs) is challenging. The difficulty lies in accurately identifying and quantifying bias and the scarcity of samples, which greatly impedes the implementation of bias correction. Current methods often adjust the distribution of presence or background points within geographic or environmental spaces to correct the sampling bias in probability estimation within SDMs. However, these methods may lead to information loss, rely on subjective assumptions, and often separate geography and environment when correcting for bias. This study proposes a novel and easily implementable method termed “aggregation background.” This method selects background data based on the aggregation degree of presence points in the geographic and environmental feature space, thereby approximating the representation and correction of sampling bias in the presence samples. We compared this new method with other prevalent sampling bias correction methods in the existing literature by analyzing ecological authenticity. Under varying biases and sample sizes, the aggregation background and geographic filtering methods achieved more accurate species distribution predictions compared to the target group background and other methods. Notably, when the sample size was small (≤70), the aggregation background was superior to that obtained using the geographic filtering method. These findings underscore the effectiveness of the aggregation background in improving bias correction using limited available presence sample data, without relying on assumptions about sampling bias. Our method provides a new approach for correcting complex unknown biases in SDMs.

Uncovering Representation Bias in Large‐scale Cellular Phone‐based Data: A Case Study in North Carolina

Large cellular phone‐based mobility datasets are an important new data source for research on human movement. We investigate and illustrate bias in representation in a large mobility data set at the census block group, tract, and county levels. We paired American Community Survey (ACS) 2019 data with SafeGraph (SG) cell phone mobility data to elucidate potential bias in SG data by examining ACS estimated population against the number of devices in the SG data, stratifying by key sociodemographic variables such as income, percent Black population, percent of population over 55 years, percent of population 18–65 years, percent of people living in crowded living conditions, and urbanization level. We evaluated whether the bias varied over time by examining a 10‐month period. This bias changes with key demographic characteristics and changes over time. Specifically, we see underrepresentation in areas that have the highest percentage of Black population at all aggregation levels. We also see underrepresentation at all levels in areas with the highest percentage of working age residents as well as areas with the lowest median incomes. Researchers should be cautious when using mobility datasets because of bias differential on key sociodemographic factors and collection time.

BayesAge: A maximum likelihood algorithm to predict epigenetic age.

Introduction: DNA methylation, specifically the formation of 5-methylcytosine at the C5 position of cytosine, undergoes reproducible changes as organisms age, establishing it as a significant biomarker in aging studies. Epigenetic clocks, which integrate methylation patterns to predict age, often employ linear models based on penalized regression, yet they encounter challenges in handling missing data, count-based bisulfite sequence data, and interpretation. Methods: To address these limitations, we introduce BayesAge, an extension of the scAge methodology originally designed for single-cell DNA methylation analysis. BayesAge employs maximum likelihood estimation (MLE) for age inference, models count data using binomial distributions, and incorporates LOWESS smoothing to capture non-linear methylation-age dynamics. This approach is tailored for bulk bisulfite sequencing datasets. Results: BayesAge demonstrates superior performance compared to scAge. Notably, its age residuals exhibit no age association, offering a less biased representation of epigenetic age variation across populations. Furthermore, BayesAge facilitates the estimation of error bounds on age inference. When applied to down-sampled data, BayesAge achieves a higher coefficient of determination between predicted and actual ages compared to both scAge and penalized regression. Discussion: BayesAge presents a promising advancement in epigenetic age prediction, addressing key challenges encountered by existing models. By integrating robust statistical techniques and tailored methodologies for count-based data, BayesAge offers improved accuracy and interpretability in predicting age from bulk bisulfite sequencing datasets. Its ability to estimate error bounds enhances the reliability of age inference, thereby contributing to a more comprehensive understanding of epigenetic aging processes.