Representativeness Research Articles

Envisioning the future of learning and teaching engineering in the artificial intelligence era: Opportunities and challenges

Envisioning the future of learning and teaching engineering in the artificial intelligence era: Opportunities and challenges

Improving the subseasonal variability of the Indian summer monsoon in a climate model

AbstractMany global climate models, including the Climate Forecast System version 2 (CFSv2), have a biased representation of subseasonal modes of variability of the Indian summer monsoon. For instance, they simulate a weaker summer mean monsoon low‐pressure systems (LPS) climatology, faster than observed northward propagation of monsoon intraseasonal oscillations (MISOs), and a systematic dry bias over Indian landmass. The Bay of Bengal (BoB), with its shallow mixed layers and unique thermal stratification, significantly modulates the convective activity in this region at subseasonal‐to‐seasonal timescales through modulation of sea surface temperature. The highly stratified upper ocean in the BoB is due to the enormous freshwater it receives from rains and rivers. A river routing model is coupled to the CFSv2 to account for the riverine freshwater and the improvements in modelling the upper‐ocean structure are analysed. Model simulations indicate that inclusion of temporally varying riverine freshwater improves the upper‐ocean state in the BoB and the observed mixed‐layer temperature gradients in the Bay are simulated reasonably after incorporating the time varying river runoff. This resulted in increased LPS lifetime and track density, and enhanced rainfall over central India. Better representation of the upper‐ocean stratification in the model leads to larger post‐convection shoaling of mixed layers at intraseasonal timescales, thereby forming thick barrier layers. Enhanced air–sea interactions restricted to the shallow mixed layer are associated with stronger vorticity, specific humidity and low‐level convergence to the north of the intraseasonal convection band. This enhanced low‐level moisture convergence north of the convection centre results in realistic northward propagation of MISO and aids LPS activity. It is demonstrated that better simulation of the upper‐ocean structure in coupled climate models can improve the representation of subseasonal modes of monsoon variability. These results bear important implications for operational forecasting.

Neural Evidence for Boundary Updating as the Source of the Repulsive Bias in Classification.

Binary classification, an act of sorting items into two classes by setting a boundary, is biased by recent history. One common form of such bias is repulsive bias, a tendency to sort an item into the class opposite to its preceding items. Sensory-adaptation and boundary-updating are considered as two contending sources of the repulsive bias, yet no neural support has been provided for either source. Here, we explored human brains of both men and women, using functional magnetic resonance imaging (fMRI), to find such support by relating the brain signals of sensory-adaptation and boundary-updating to human classification behavior. We found that the stimulus-encoding signal in the early visual cortex adapted to previous stimuli, yet its adaptation-related changes were dissociated from current choices. Contrastingly, the boundary-representing signals in the inferior-parietal and superior-temporal cortices shifted to previous stimuli and covaried with current choices. Our exploration points to boundary-updating, rather than sensory-adaptation, as the origin of the repulsive bias in binary classification.SIGNIFICANCE STATEMENT Many animal and human studies on perceptual decision-making have reported an intriguing history effect called "repulsive bias," a tendency to classify an item as the opposite class of its previous item. Regarding the origin of repulsive bias, two contending ideas have been proposed: "bias in stimulus representation because of sensory adaptation" versus "bias in class-boundary setting because of belief updating." By conducting model-based neuroimaging experiments, we verified their predictions about which brain signal should contribute to the trial-to-trial variability in choice behavior. We found that the brain signal of class boundary, but not stimulus representation, contributed to the choice variability associated with repulsive bias. Our study provides the first neural evidence supporting the boundary-based hypothesis of repulsive bias.

Cognitive, clinical and genetic characterization of dementia in Cali, Colombia: A pilot study in a diverse population

AbstractBackgroundThe aging population and the prevalence of dementia are rapidly growing in Latin America. People with dementia are underdiagnosed and not well characterized in this region. The scenario is complicated by limited options for culturally proper testing and Latinos' socio‐biological and phenotypic diversity, which affect dementia presentation. In Cali, frontotemporal dementia (FTD) is misdiagnosed as a psychiatric disorder in >70% of cases. Delay in the accurate diagnosis and misdiagnosis of dementia affects genetic counseling, treatment, prognosis, and caregiver burden. Furthermore, most of the discovery efforts in genetics are based on data from populations of European ancestry. This bias in representation can exacerbate existing disease and healthcare disparities. Innovative technology for neuropsychological assessment could help make an early and accurate diagnosis. Genetic analysis is novel in this region, and the proposed work lays the basis for the genomic studies we advocate for.MethodWe will evaluate the accuracy of the TabCAT cognitive tests for the identification of cognitive impairment in participants previously diagnosed with FTD (N = 20), AD (N = 20), and healthy controls (N = 40). We also will assess the NPI. In addition, we will collect family history information for neurodegenerative diseases, genome sequencing of the sample.ResultWe hypothesize that the TabCAT evaluation will show sensitivity and specificity >.80 for detecting cognitive impairment, exceeding the accuracy of standard cognitive screenings, such as the Montreal Cognitive Assessment (MoCA). The assessment will be similarly accurate for detecting cognitive impairment in AD and bvFTD, each separately compared to controls. For participants with bvFTD, the NPI will improve the classification of bvFTD from controls and AD, above and beyond the cognitive tests. We hypothesize that 10‐20% of the participants will have a family history of dementia and be identified with genetic risk for neurodegenerative diseases.ConclusionThis pilot lays the groundwork to establish a new site for ReDLat, a consortium of multiple partners to expand dementia research in Latin America and the Caribbean. Assessing this underrepresented population, we will contribute to an accurate characterization of dementia in Colombia, which will improve dementia diagnosis and reduce health inequities in this region.

Unbiased Spatiotemporal Representation With Uncertainty Control for Person Reidentification

For person re-identification, most current research aims to encode the spatial and temporal information by using convolutional neural networks (CNNs) to extract spatial features and recurrent neural networks (RNNs) or their variations to discover the time dependencies. However, it ignores the effect of the complex background, which leads to a biased spatial representation. Further, it often uses the backpropagation through time (BPTT) to train RNNs. Unfortunately, it is hard to learn the long-term dependency via BPTT due to the gradient vanishing or exploding. The significance of a frame should not be biased by its position in a given sequence. In this paper, a new method is proposed to learn an unbiased semantic representation for video-based person re-identification. To handle the background clutter and occlusion, a two-branch CNN model is used to obtain the enriched representation from both the foreground person and original pedestrian images. Then, an unbiased bidirectional convolutional neural network architecture is developed to learn the unbiased spatial and temporal representation. Experimental results on three public datasets demonstrate the effectiveness of the proposed method.

Dual Instance-Consistent Network for Cross-Domain Object Detection.

Cross-domain object detection aims to transfer knowledge from a labeled dataset to an unlabeled dataset. Most existing methods apply a unified embedding model to generate the tightly coupled source and target descriptions for domain alignment, leading to the destroyed feature distribution of the target domain because the embedding model is mainly controlled by the source domain. To reduce the representation bias of the target domain, we apply two independent networks to extract two types of discriminative descriptions with mutual consistency, i.e., a novel Dual Instance-Consistent Network (DICN) is proposed for cross-domain object detection. Especially, Dual Instance-Consistent Module containing the instance mutual consistency between Primary Network and Auxiliary Network is applied to align two domains, where Primary and Auxiliary Networks are used to obtain the source-specific and target-specific information, respectively. The instance mutual consistency consists of two terms: feature consistency and detection consistency, which is applied to align the instance feature and the output of detection head, respectively. With the instance mutual consistency, optimizing the Primary (Auxiliary) Network only with source (target) images by fixing the Auxiliary (Primary) Network can generate the source(target)-specific description. Extensive experiments on several benchmarks demonstrate the effectiveness of the proposed DICN, e.g., obtaining mAP of 44.10% for Cityscapes → Foggy Cityscapes, AP on car of 76.50% for Cityscapes → KITTI, MR -2 of 8.87%, 12.66%, 22.27%, and 42.06% for COCOPersons → Caltech, CityPersons → Caltech, COCOPersons → CityPersons, and Caltech → CityPersons, respectively.

Colorectal polyp segmentation method based on fusion of transformer and cross-level phase awareness

In order to address the issues of spatial induction bias and lack of effective representation of global contextual information in colon polyp image segmentation, which lead to the loss of edge details and mis-segmentation of lesion areas, a colon polyp segmentation method that combines Transformer and cross-level phase-awareness is proposed. The method started from the perspective of global feature transformation, and used a hierarchical Transformer encoder to extract semantic information and spatial details of lesion areas layer by layer. Secondly, a phase-aware fusion module (PAFM) was designed to capture cross-level interaction information and effectively aggregate multi-scale contextual information. Thirdly, a position oriented functional module (POF) was designed to effectively integrate global and local feature information, fill in semantic gaps, and suppress background noise. Fourthly, a residual axis reverse attention module (RA-IA) was used to improve the network's ability to recognize edge pixels. The proposed method was experimentally tested on public datasets CVC-ClinicDB, Kvasir, CVC-ColonDB, and EITS, with Dice similarity coefficients of 94.04%, 92.04%, 80.78%, and 76.80%, respectively, and mean intersection over union of 89.31%, 86.81%, 73.55%, and 69.10%, respectively. The simulation experimental results show that the proposed method can effectively segment colon polyp images, providing a new window for the diagnosis of colon polyps.

Semantic Web technologies and bias in artificial intelligence: A systematic literature review

Bias in Artificial Intelligence (AI) is a critical and timely issue due to its sociological, economic and legal impact, as decisions made by biased algorithms could lead to unfair treatment of specific individuals or groups. Multiple surveys have emerged to provide a multidisciplinary view of bias or to review bias in specific areas such as social sciences, business research, criminal justice, or data mining. Given the ability of Semantic Web (SW) technologies to support multiple AI systems, we review the extent to which semantics can be a “tool” to address bias in different algorithmic scenarios. We provide an in-depth categorisation and analysis of bias assessment, representation, and mitigation approaches that use SW technologies. We discuss their potential in dealing with issues such as representing disparities of specific demographics or reducing data drifts, sparsity, and missing values. We find research works on AI bias that apply semantics mainly in information retrieval, recommendation and natural language processing applications and argue through multiple use cases that semantics can help deal with technical, sociological, and psychological challenges.

Use of stacked Gaussian processes regression method to improve prediction of groundwater solute transport model

The complex groundwater solute transport model can generally be divided into several coupled submodels, the systematic error is inevitable in each submodel due to the unreasonable simplification or biased representation of the true groundwater environment system. The submodels’ systematic errors and their propagation lead to great challenge for reliable model predictions. To improve the predictive performance of groundwater solute transport model, the Stacked Gaussian processes regression (GPR) method was proposed to quantify and reduce the error propagation among the submodels in this study. Stacked GPR corrects the structure errors of groundwater model submodel by submodel gradually. Two case studies are performed to demonstrate the effectiveness of Stacked GPR method, including a synthetical groundwater transport case and a laboratory NaCl solute transport experiment. The results demonstrated that Stacked GPR can reduce the error propagation among submodels and improve model’s predictive performance apparently. Moreover, it is more feasible to correct the structure error of each submodel in sequence by Stacked GPR than to correct the lumped structure error by traditional GPR for a component-based groundwater solute transport model. The errors in prior submodels could aggravate the complexity of the error of the posterior submodel though error propagation, which increases the difficulty of constructing an effective error-correcting model.

Are Searches in OCR-generated Archives Trustworthy?

Abstract Digitised archives are revolutionary tools for research that, in a few seconds, generate results that earlier often took years to obtain. But do they provide all results for the terms searched for? The accuracy of searches was tested by performing sample searches of leading newspaper databases. The test revealed several weaknesses in the search process, including an average 18 percent error rate for single words in body text, and a far higher error rates for advertisements. Such high error rates encourage a critical look at the 20-year-old sector. Although these errors can be reduced by a re-digitation and with new improved OCR engines and new search algorithms, searches will nevertheless return manipulated results. In response, and to identify infringed bias and skewed representation, database owners need to provide thorough metadata to ensure source criticism.

Queer Identities, Normative Databases: Challenges to Capturing Queerness On Wikidata

The collection, organization, and retrieval of data about queer individuals and their identities challenge the creators and curators of highly structured database systems. While prior research in archival studies and demographics has examined processes of collecting and storing queer identities, they do not examine the complexities of highly democratized platforms that lack top-down mandates that often structure archival schemas. To examine the representation of queer people on open-platform databases, we performed a trace ethnography and thematic analysis of Wikidata, an open collaboration, highly structured database. We specifically examined the creation of, changes to, discussions around, and impacts of properties that encode queer identities, such assexual orientation andsex or gender. We found that changes often have unexpected impacts, that contributors struggled to determine vocabulary for queer identities which were accurate across the diverse cultural contexts of the Wikidata community, that the recording of queer identities could cause a stigmatizing effect for LGBTQ+ individuals, with further concerns of spreading rumors or outing closeted people, and that contributors proposing changes which would cause biased representations of queer people. Our analysis demonstrates inherent and unaddressed frictions when translating queer identities to the confines of a structured database. We conclude by discussing ways that the highly bottom-up, collaborative nature of platforms such as Wikidata, often seen as a major strength, can be vulnerable to individuals or small groups derailing and filibustering changes they disagree with on politically charged topics such as queer identities.

Quantitative analysis of tRNA abundance and modifications by nanopore RNA sequencing

Transfer RNAs (tRNAs) play a central role in protein translation. Studying them has been difficult in part because a simple method to simultaneously quantify their abundance and chemical modifications is lacking. Here we introduce Nano-tRNAseq, a nanopore-based approach to sequence native tRNA populations that provides quantitative estimates of both tRNA abundances and modification dynamics in a single experiment. We show that default nanopore sequencing settings discard the vast majority of tRNA reads, leading to poor sequencing yields and biased representations of tRNA abundances based on their transcript length. Re-processing of raw nanopore current intensity signals leads to a 12-fold increase in the number of recovered tRNA reads and enables recapitulation of accurate tRNA abundances. We then apply Nano-tRNAseq to Saccharomyces cerevisiae tRNA populations, revealing crosstalks and interdependencies between different tRNA modification types within the same molecule and changes in tRNA populations in response to oxidative stress.

Multivariate Drought Monitoring, Propagation, and Projection Using Bias‐Corrected General Circulation Models

AbstractUnderstanding how droughts are characterized, propagated, and projected, particularly multivariate droughts, is necessary to explain the variability and changes in drought characteristics. This study aims to understand multimodel global drought monitoring, propagation, and projection by utilizing a multivariate standardized drought index (MSDI) during the historical (1959–2014) and future (2045–2100) periods under two socioeconomic pathways SSPs (370 and 585), derived from the bias‐corrected Coupled Model Intercomparison Project Phase 6 (CMIP6). Based on the energy metrics, the multivariate bias correction method outperformed other techniques in correcting the biases in the CMIP6 drought representation. The drought indicators demonstrate distinct categories for meteorological, hydrological, and multivariate droughts. There were significant high cross correlations between Heatwave Total Length (HWTL) and MSDI in Africa and South America for all lagged times. Europe and North America generally saw the maximum MSDI drought duration (228 months) during the historical period. For future projections, Africa recorded the maximum drought duration (197 months), while Europe witnessed the minimum drought duration for SSP 370 (171 months), and North America (149 months) for SSP 585. Furthermore, during the historical period in tropical Africa, the propagation of meteorological to hydrological drought was slower during the wet months than during the dry months. Under the SSP 370 future projection, there was a shift in the long period of meteorological‐hydrological propagation from the middle and late wet months to the beginning of the wet months in tropical Africa. Therefore, tracking and projecting drought characteristics is vital for understanding the risk of drought‐related consequences.

조세불복사건 판단에 있어 인지 편향과 극복방안

Humans acquired heuristics for survival. Heuristics are mental shortcuts that save time by simplifying complex decisions into simple rules of thumb. Heuristics are useful in everyday life because they enable us to automatically make quick decisions with little effort. However, when making important decisions and legal judgments, heuristics can be dangerous. Before filing a lawsuit in an appeals proceeding against a taxation, it must be needed to file an administrative appeal. It is the procedure enforced by law. Administrative hearings is similar to the judicial proceedings. The evaluating of the weight of proof depends on the committee members just as the judge. The problem is that the judge could make wrong decision due to his or her cognitive bias. In the United States, many experimental studies have been conducted on jurors and judges to see if various heuristics actually occur. It is important for parties to accept the legal decision, so that the legal decision must not be biased.
 The purpose of this research is to analyze the cases that had been shown heuristics in administrative trial decisions and to research on debiasing. This paper analyzes decisions in which representativeness heuristics, hindsight bias, and confirmation bias appeared. In order to debiase, there are several trials like the cognitive correction(thinking the opposite etc.), environmental correction(reducing cognitive losses etc.), and systemic correction(decision support systems etc.).

Association of Demographic and Socioeconomic Indicators With the Use of Wearable Devices Among Children

The use of consumer-grade wearable devices for collecting data for biomedical research may be associated with social determinants of health (SDoHs) linked to people's understanding of and willingness to join and remain engaged in remote health studies. To examine whether demographic and socioeconomic indicators are associated with willingness to join a wearable device study and adherence to wearable data collection in children. This cohort study used wearable device usage data collected from 10 414 participants (aged 11-13 years) at the year-2 follow-up (2018-2020) of the ongoing Adolescent Brain and Cognitive Development (ABCD) Study, performed at 21 sites across the United States. Data were analyzed from November 2021 to July 2022. The 2 primary outcomes were (1) participant retention in the wearable device substudy and (2) total device wear time during the 21-day observation period. Associations between the primary end points and sociodemographic and economic indicators were examined. The mean (SD) age of the 10 414 participants was 12.00 (0.72) years, with 5444 (52.3%) male participants. Overall, 1424 participants (13.7%) were Black; 2048 (19.7%), Hispanic; and 5615 (53.9%) White. Substantial differences were observed between the cohort that participated and shared wearable device data (wearable device cohort [WDC]; 7424 participants [71.3%]) compared with those who did not participate or share data (no wearable device cohort [NWDC]; 2900 participants [28.7%]). Black children were significantly underrepresented (-59%) in the WDC (847 [11.4%]) compared with the NWDC (577 [19.3%]; P < .001). In contrast, White children were overrepresented (+132%) in the WDC (4301 [57.9%]) vs the NWDC (1314 [43.9%]; P < .001). Children from low-income households (<$24 999) were significantly underrepresented in WDC (638 [8.6%]) compared with NWDC (492 [16.5%]; P < .001). Overall, Black children were retained for a substantially shorter duration (16 days; 95% CI, 14-17 days) compared with White children (21 days; 95% CI, 21-21 days; P < .001) in the wearable device substudy. In addition, total device wear time during the observation was notably different between Black vs White children (β = -43.00 hours; 95% CI, -55.11 to -30.88 hours; P < .001). In this cohort study, large-scale wearable device data collected from children showed considerable differences between White and Black children in terms of enrollment and daily wear time. While wearable devices provide an opportunity for real-time, high-frequency contextual monitoring of individuals' health, future studies should account for and address considerable representational bias in wearable data collection associated with demographic and SDoH factors.

General Words Representation Method for Modern Language Model

This paper proposes a new word representation method emphasizes general words over specific words. The main motivation for developing this method is to address the weighting bias in modern Language Models (LMs). Based on the Transformer architecture, contemporary LMs tend to naturally emphasize specific words through the Attention mechanism to capture the key semantic concepts in a given text. As a result, general words, including question words are often neglected by LMs, leading to a biased word significance representation (where specific words have heightened weight, while general words have reduced weights). This paper presents a case study, where general words' semantics are as important as specific words' semantics, specifically in the abstractive answer area within the Natural Language Processing (NLP) Question Answering (QA) domain. Based on the selected case study datasets, two experiments are designed to test the hypothesis that "the significance of general words is highly correlated with its Term Frequency (TF) percentage across various document scales”. The results from these experiments support this hypothesis, justifying the proposed intention of the method to emphasize general words over specific words in any corpus size. The output of the proposed method is a list of token (word)-weight pairs. These generated weights can be used to leverage the significance of general words over specific words in suitable NLP tasks. An example of such task is the question classification process (classifying question text whether it expects factual or abstractive answer). In this context, general words, particularly the question words are more semantically significant than the specific words. This is because the same specific words in different questions might require different answers based on their question words (e.g. "How many items are on sale?" and "What items are on sale?" questions). By employing the general weight values produced by this method, the weightage of question and specific words can be heightened, making it easier for the classification system to differentiate between these questions. Additionally, the token (word)-weight pair list is made available online at https://www.kaggle.com/datasets/saliimiabbas/genwords-weight.

Estimating fertility using adults: A method for under‐enumerated pre‐adult skeletal samples

Infant underrepresentation poses a great risk to accurate palaeodemographic findings when analyzing skeletal samples. Empirically derived palaeodemographic methods all require unbiased or minimally biased pre-adult representation for estimating demographic characteristics, including fertility. Currently, there are no reliable methods for estimating palaeodemographic parameters when pre-adults are underrepresented in skeletal samples, consequently such samples are often excluded from palaeodemographic analyses. The aim of this article is to develop a method for estimating total fertility rate (TFR) using reproductive aged adults, specifically for samples with suspected pre-adult under-enumeration. United Nations mortality data and TFR from the World Population Prospects was utilized. The correlation between known fertility and the proportion of individuals in key reproductive years (15-49 years) to total adult sample (15+ years) was assessed as an indirect means to estimate fertility. It was determined that the proportion of reproductive aged adults is a reasonable proxy for fertility. A significant positive correlation was observed between the TFR and those who died aged 15-49 years of age as a proportion of those who died ≥15 years (D15-49/D15+). SE of the estimate revealed reasonable predictive accuracy. When applied to two modern non-agricultural populations, the method showed some variability in accuracy but good potential for an improved outcome over existing methods when pre-adults are underrepresented. This research has provided a new method for estimating fertility in archeological skeletal samples with pre-adult under-enumeration. In combination with a contextually focused approach, this provides a significant step toward further use of biased samples in palaeodemography.

Comparing Vaginal and Endometrial Microbiota Using Culturomics: Proof of Concept.

It is generally accepted that microorganisms can colonize a non-pathological endometrium. However, in a clinical setting, endometrial samples are always collected by passing through the vaginal-cervical route. As such, the vaginal and cervical microbiomes can easily cross-contaminate endometrial samples, resulting in a biased representation of the endometrial microbiome. This makes it difficult to demonstrate that the endometrial microbiome is not merely a reflection of contamination originating from sampling. Therefore, we investigated to what extent the endometrial microbiome corresponds to that of the vagina, applying culturomics on paired vaginal and endometrial samples. Culturomics could give novel insights into the microbiome of the female genital tract, as it overcomes sequencing-related bias. Ten subfertile women undergoing diagnostic hysteroscopy and endometrial biopsy were included. An additional vaginal swab was taken from each participant right before hysteroscopy. Both endometrial biopsies and vaginal swabs were analyzed using our previously described WASPLab-assisted culturomics protocol. In total, 101 bacterial and two fungal species were identified among these 10 patients. Fifty-six species were found in endometrial biopsies and 90 were found in vaginal swabs. On average, 28 % of species were found in both the endometrial biopsy and vaginal swab of a given patient. Of the 56 species found in the endometrial biopsies, 13 were not found in the vaginal swabs. Of the 90 species found in vaginal swabs, 47 were not found in the endometrium. Our culturomics-based approach sheds a different light on the current understanding of the endometrial microbiome. The data suggest the potential existence of a unique endometrial microbiome that is not merely a presentation of cross-contamination derived from sampling. However, we cannot exclude cross-contamination completely. In addition, we observe that the microbiome of the vagina is richer in species than that of the endometrium, which contradicts the current sequence-based literature.

Deep attention models with dimension-reduction and gate mechanisms for solving practical time-dependent vehicle routing problems

Time dependencies of travel speeds in time-dependent vehicle routing problems (TDVRPs) are usually accounted for by discretizing the planning horizon into several time periods. However, travel speeds usually change frequently in real road networks, so many time periods are needed to evaluate candidate solutions accurately in model and solution construction for practical TDVRPs, which increases substantially the computational complexity of TDVRPs. We develop two deep attention models with dimension-reduction and gate mechanisms to solve practical TDVRPs in real urban road networks. In the two models, a multi-head attention-based dimension-reduction mechanism is proposed to reduce the dimension of model inputs and obtain enhanced node representation, whereas a gate mechanism is introduced to obtain better information representation. On the basis of a travel speed dataset from an urban road network, we conduct extensive experiments to validate the effectiveness of the proposed models on practical TDVRPs with or without consideration of time windows. Experimental results show that our models can solve TDVRPs with 240 time periods and up to 250 customers effectively and efficiently and provide significantly superior overall performances over two representative heuristics and two state-of-the-art deep reinforcement learning models. Especially, compared with a recent tabu search method, our models can reduce the computation time by up to 3,540 times and improve the solution performance by up to 23%. Moreover, our models have an outstanding generalization performance. The model trained for the 30-customer TDVRP with time windows can be used directly to solve problems with up to 250 customers effectively by generating superior solutions over those generated by benchmarking methods.

Perceptual history biases in serial ensemble representation.

Ensemble perception refers to the visual system's ability to efficiently represent groups of similar objects as a unified percept using their summary statistical information. Most studies focused on extraction of current trial averages, giving little attention to prior experience effects, although a few recent studies found that ensemble mean estimations contract toward previously presented stimuli, with most of these focusing on explicit perceptual averaging of simultaneously presented item ensembles. Yet, the time element is crucial in real dynamic environments, where we encounter ensemble items over time, aggregating information until reaching summary representations. Moreover, statistical information of objects and scenes is learned over time and often implicitly and then used for predictions that shape perception, promoting environmental stability. Therefore, we now focus on temporal aspects of ensemble statistics and test whether prior information, beyond the current trial, biases implicit perceptual decisions. We designed methods to separate current trial biases from those of previously seen trial ensembles. In each trial, six circles of different sizes were presented serially, followed by two test items. Participants were asked to choose which was present in the sequence. Participants unconsciously rely on ensemble statistics, choosing stimuli closer to the ensemble mean. To isolate the influence of earlier trials, the two test items were sometimes equidistant from the current trial mean. Results showed membership judgment biases toward current trial mean, when informative (largest effect). On equidistant trials, judgments were biased toward previously experienced stimulus statistics. Comparison of similar conditions with a shifted stimulus distribution ruled out a bias toward an earlier, presession, prototypical diameter. We conclude that ensemble perception, even for temporally experienced ensembles, is influenced not only by current trial mean but also by means of recently seen ensembles and that these influences are somewhat correlated on a participant-by-participant basis.