Prior Information Research Articles

Self-supervised learning fabric defect segmentation using anomaly generation

PurposeThe problem of fabric defect detection is a particularly challenging task, as the fabric defects occupy only a small portion of the image pixels and it is difficult to collect sufficient samples for training deep learning-based models. The purpose of this work is to present a novel self-supervised learning method to address this problem.Design/methodology/approachIn order to solve the problem of lack of defect samples, based on the fabric-specific degree of texture regularity, we propose an anomaly generation method to create synthetic fabric defects by destroying the normal fabric texture. To improve the detection of defects of different sizes, a global–local parallel detection mechanism is proposed. A self-supervised model including an anomaly generation module, a reconstruction subnetwork and a discriminative subnetwork is established to achieve model training without prior anomaly information.FindingsThe proposed method features self-supervised training, does not require no labelled anomaly data and detects anomalies by distinguishing their distance from normal samples at both global and local levels. When tested on four fabric datasets, our approach outperforms state-of-the-art unsupervised and self-supervised methods and achieves significantly higher localization accuracy.Originality/valueA high-fidelity fabric defects synthesis method is presented to alleviate the problem of collecting numerous fabric defects, providing a reference for other surface anomaly detection problems. The proposed global–local parallel detection mechanism can serve as a reference for other methods dedicated to detecting particular anomalies. The proposed self-supervised network model can effectively locate fabric anomalies without prior labelling information, which could be used as a framework for other model designs.

Deep 3D-DIC using a coarse-to-fine network for robust and accurate 3D shape and displacement measurements

Deep learning has become an attractive tool for addressing the limitations of traditional digital image correlation (DIC). However, extending learning-based DIC methods to three-dimensional (3D-DIC) measurements is challenging due to the limited displacement estimation range, which cannot handle the large displacements caused by stereo-matching disparities. Besides, most of the existing learning-based DIC architectures lack prior information to guide displacement estimation, resulting in insufficient accuracy. To solve these problems, we proposed a learning-based 3D-DIC (i.e., Deep 3D-DIC) using a coarse-to-fine network called G-RAFT for large and accurate image displacement estimation. Specifically, the large displacement estimation network GMA is adopted to calculate the large coarse displacement field, which is further warped on the deformed image to eliminate the main displacement component. The residual small deformation between the reference image and the warped image is further extracted using the recently proposed RAFT-DIC with high accuracy. By subtracting small displacement from large displacement, the refined displacement field is obtained. In contrast to standard subset-based 3D-DIC, Deep 3D-DIC achieves full-automatic pixel-wise 3D shape and displacement reconstruction without manual parameter input. Experimental results demonstrate that Deep 3D-DIC achieves accuracy comparable to subset-based 3D-DIC, with strong generalization ability and remarkable advantages in scenarios with complex surfaces.

On Loopy Belief Propagation for SASCAs

Profiled power analysis is one of the most powerful forms of passive side-channel attacks. Over the last two decades, many works have analyzed their impact on cryptographic implementations as well as corresponding countermeasure techniques. To date, the most advanced variants of profiled power analysis are based on Soft-analytical Side-Channel Attacks (SASCA). After the initial profiling phase, a SASCA adversary creates a probabilistic graphical model, called a factor graph, of the target implementation and encodes the results of the previous step as prior information. Then, an inference algorithm such as loopy Belief Propagation (BP) can be used to recover the distribution of a target variable in the graph, i.e., sensitive data/keys. Designers of cryptographic implementations aim to reduce information leakage as much as possible and assess how much leakage can be allowed without compromising security requirements. Despite the existence of many works on profiled power analysis, it is still notoriously difficult to state under which conditions a cryptographic implementation provides sufficient protection against a profiling attacker with certain capabilities. In particular, it is unknown when a BP-based attack is optimal or whether tuning some heuristics in that algorithm may significant strengthen the attack. This knowledge gap led us to investigate the effectiveness of BP for SASCAs by studying the modes of failures of BP in the context of the SASCA, and systematically analyzing the behavior of BP on practically-relevant factor graphs. We use exact inference to gauge the quality of the approximation provided by BP. Through this assessment, we show that there exists a significant disparity between BP and exact inference in terms of guessing entropy when performing SASCAs on several classes of factor graphs. We further review and analyze various BP improvement heuristics from the literature.

Quantifying Uncertainty of Insurance Claims Based on Expert Judgments

In Bayesian statistics, prior specification has an important role in determining the quality of posterior estimates. We use expert judgments to quantify uncertain quantities and produce appropriate prior distribution. The aim of this study was to quantify the uncertainty of life insurance claims, especially on the policy owner’s age, as it is the main factor determining the insurance premium. A one-day workshop was conducted to elicit expert judgments from those who have experience in accepting claims. Four experts from different insurance companies were involved in the workshop. The elicitation protocol used in this study was The Sheffield Elicitation Framework (SHELF), which produces four different statistical distributions for each expert. A linear pooling method was used to aggregate the distributions to obtain the consensus distribution among experts. The consensus distribution suggested that the majority of policy owners will make a claim at the age of 54 years old.

PARAMETER ESTIMATION OF LOGNORMAL AND PARETO TYPE I DISTRIBUTIONS USING FREQUENTIST AND BAYESIAN INFERENCES

Extreme events are events that rarely occur but they cause substantial losses. Insurance companies need to take extreme events into account in risk management because extreme events can have a negative impact on the company's financial health. As a result, insurance companies need an appropriate loss model that matches the empirical data from these extreme events. A distribution that is heavy-tailed and skewed to the right is a good distribution for modeling the magnitude of losses from extreme events. In this paper, two distributions with heavy tails and skew to the right will be used to model the magnitude of losses from extreme events, namely the lognormal distribution and the Pareto distribution type I. The parameters of these distributions are estimated using two inferences, namely the frequentist and Bayesian inferences. In the frequentist inference, two methods are applied, namely the moment method and maximum likelihood. On Bayesian inference, two prior distributions are used, namely uniform and Jeffrey. Test model suitability is carried out by visually comparing the model distribution function with the empirical distribution function, as well as by comparing the Root Mean Square Error (RMSE) value. The visualization results of the distribution function and RMSE values ​​show that in general, the Bayesian inference is better at estimating parameters than the frequentist inference. In the frequentist inference, the maximum likelihood method can provide better estimated values ​​than the moment method. In the Bayesian inference, the two prior distributions show a relatively similar fit to the data and tend to be better than the frequentist inference.

Genetic Algorithm-Enhanced Direct Method in Protein Crystallography

Direct methods based on iterative projection algorithms can determine protein crystal structures directly from X-ray diffraction data without prior structural information. However, traditional direct methods often converge to local minima during electron density iteration, leading to reconstruction failure. Here, we present an enhanced direct method incorporating genetic algorithms for electron density modification in real space. The method features customized selection, crossover, and mutation strategies; premature convergence prevention; and efficient message passing interface (MPI) parallelization. We systematically tested the method on 15 protein structures from different space groups with diffraction resolutions of 1.35∼2.5 Å. The test cases included high-solvent-content structures, high-resolution structures with medium solvent content, and structures with low solvent content and non-crystallographic symmetry (NCS). Results showed that the enhanced method significantly improved success rates from below 30% to nearly 100%, with average phase errors reduced below 40°. The reconstructed electron density maps were of sufficient quality for automated model building. This method provides an effective alternative for solving structures that are difficult to predict accurately by AlphaFold3 or challenging to solve by molecular replacement and experimental phasing methods. The implementation is available on Github.

Bayesian Point Estimation of Linear/Circular Consecutive k‐out‐of‐n : F System Reliability Under Vague Environment

The paper proposes a Bayesian approach to estimate the reliability and mean time to failure (MTTF) of linear and circular consecutive [Formula: see text] systems under vague environments, where uncertainties are represented using vague sets. By assuming model parameters as vague random variables with vague prior distributions, the study employs the Resolution Identity theorem for vague sets to resolve and structure this vagueness within the Bayesian framework. The proposed method uses the squared error loss function to derive vague Bayesian estimates, providing a more realistic and comprehensive analysis of system reliability in the presence of uncertainty. Numerical illustrations are included to demonstrate the applicability and effectiveness of the approach, highlighting its potential for handling vague information in complex system reliability analysis.

Multi-Centroid Extraction Method for High-Dynamic Star Sensors Based on Projection Distribution of Star Trail

To improve the centroid extraction accuracy and efficiency of high-dynamic star sensors, this paper proposes a multi-centroid localization method based on the prior distribution of star trail projections. First, the mapping relationship between attitude information and star trails is constructed based on a geometric imaging model, and an endpoint centroid group extraction strategy is designed from the perspectives of time synchronization and computational complexity. Then, the endpoint position parameters are determined by fitting the star trail grayscale projection using a line spread function, and accurate centroid localization is achieved through principal axis analysis and inter-frame correlation. Finally, the effectiveness of the proposed method under different dynamic scenarios was tested using numerical simulations and semi-physical experiments. The experimental results show that when the three-axis angular velocity reaches 8°/s, the centroid extraction accuracy of the proposed method remains superior to 0.1 pixels, achieving an improvement of over 30% compared to existing methods and simultaneously doubling the attitude measurement frequency. This demonstrates the superiority of this method in high-dynamic attitude measurement tasks.

Objective Bayesian analysis of common quantiles of normal distributions

. Many procedures can assess the equivalence of the means or medians between groups when comparing several independent populations. However, neither the mean nor median captures the entire distribution. We propose a Bayesian approach for the quantiles of several normal distributions, focusing on developing non informative priors to compare the quantiles of multiple populations. Further, reference priors and first- and second-order matching priors are developed. The second-order matching prior does not exist, and the reference priors do not satisfy a first-order matching criterion. In addition, conditions for proper posterior distributions for general priors, including developed priors, are provided. Finally, simulations include a real-world example to demonstrate the proposed approach. The numerical studies indicate that the matching prior performs better than the reference prior and Jeffreys’ prior in terms of matching the frequentist target coverage probabilities, and shows good performance even in small sample sizes. Moreover, the matching prior is not affected by the values of parameters, the number of populations, and does not depend on the values of pth quantile

An introduction to Bayesian geospatial analysis using a Bayesian multilevel model case study

ABSTRACT Despite the rapidly growing interest in Bayesian inference, recent publications reveal common issues that suggest even experienced researchers across disciplines, including geospatial research, may not always follow the proper procedure in conducting Bayesian analysis and report results. This study aims to promote Bayesian inference and the best practice guidelines of Bayesian analysis, targeting beginners in the geospatial community. We selected the Bayesian multilevel model (BMLM) to demonstrate proper Bayesian analysis through a case study examining the effect of neighbourhood-level social deprivation on birthweight in Fulton County, Georgia, USA. We constructed both frequentist multilevel models (MLMs) and BMLMs, and these were two-level varying intercept models. Following the Avoid the Misuse of Bayesian Statistics (WAMBS) checklist, we illustrated the BMLM workflow and highlighted key steps in fitting, reporting, and interpreting BMLM. Our case study shows several advantages of the Bayesian approach over the frequentist method, including incorporating prior information, better handling of uncertainty, more intuitive interpretation, and greater transparency in reporting. However, some benefits, such as avoiding multiple comparisons and generating robust estimations, are challenging to explicitly illustrate. Other benefits such as handling small sample sizes and complex models cannot be showcased with our simple models. Notably, BMLM is more computationally intensive compared to MLMs. Adhering to established guidelines can enhance the quality, transparency, and reproducibility of Bayesian analysis, allowing us to fully harness the potential of Bayesian inference in advancing geospatial knowledge.

Towards estimating the proportion of dead and missing vines at the field level

This work aimed to improve practices for estimating dead and missing vines, particularly when these estimations are made via sampling. Vine mortality is a well-known phenomenon that many studies have attempted to better explain, but only a few have focused on estimating the number of dead and missing vines at the field-level, which remains a challenging task for grapegrowers. The present article aimed to determine to what extend an increased an increased sampling effort improves estimation accuracy and thus help practitioners define a sample size adapted to vineyard properties and the accuracy expected by the grower. The first part of the study investigated whether vineyard properties (year of planting and variety) and available ancillary data (soil resistivity and fraction of vegetation cover [Fcover]) can provide a priori information on the proportion of dead and missing vines and guide the sampling strategy. The analysis was based on an exhaustive dataset created by individually mapping 14,199 dead and missing vines across 29 fields in a 20 ha vineyard. In this vineyard, regression models showed an increase in dead and missing vines of around 2.2 % per year from the tenth year onwards. Plantation year, variety and vigour (assessed by Fcover) were identified as providing valuable prior information on the proportion of dead and missing vines within the fields. The second part of the study focused on characterising the estimation error values that can result from these estimations. It focuses on sampling estimation, highlighting to what extent field properties (such as the number of dead and missing vines and their spatial autocorrelation) affect the accuracy of estimates. The results showed that the same sampling protocol can result in different accuracies from one field to another. Thus, they underline the importance of leveraging prior information, such as available field data, to tailor sampling efforts accordingly. Based on these results, this article introduces new guidelines to facilitate the estimation of the proportion of dead and missing vines in vineyards.

Development of Bayesian segmented Poisson regression model to forecast COVID-19 dynamics based on wastewater data: a case study in Nanning City, China

IntroductionCOVID-19 has caused tremendous hardships and challenges around the globe. Due to the prevalence of asymptomatic and pre-symptomatic carriers, relying solely on disease testing to screen for infections is not entirely reliable, which may affect the accuracy of predictions about the pandemic trends. This study is dedicated to developing a predictive model aimed at estimating of the dynamics of COVID-19 at an early stage based on wastewater data, to assist in establishing an effective early warning system for disease control.MethodViral load in wastewater and the number of daily reported COVID-19 cases were collected from Nanning CDC and the Chinese Disease Prevention and Control Information System, respectively. We used the viral load to estimate daily reported cases by a Bayesian linear regression model. Subsequently, a Bayesian (segmented) Poisson regression model was developed, using data from the first wave of the epidemic as prior information, to predict the COVID-19 epidemic trend of the second wave. Finally, in order to explore the optimal training data for predicting outbreak dynamics during the pandemic, we fitted the model using various training sets.ResultsThe results revealed the estimated cases, using the viral load with a 3-day lag, were consistent with the actual reported cases, with adjusted R² value of 0.935 (p < 0.001). Our model successfully predicted the epidemic peak time and provided early warnings on the third day after the outbreak began. Furthermore, after using data from the first 6 days of the outbreak, the model’s MAPE rapidly decreasing to lower levels (MAPE = 29.34%) and eventually stabilized at approximately 20%. Compared to using non-informative priors, this result allows for an advance warning of approximately two weeks. Importantly, as the inclusion of data from early outbreak increased, the predictive results of the model became more stable and accurate.ConclusionThis study demonstrates the potential of wastewater-based epidemiology combined with Bayesian methods as a monitoring and predictive tool during infectious disease outbreaks.

Bayesian item response theory to estimate power in clinical trials with patient-reported outcomes as endpoints.

Patient-Reported Outcomes (PROs) are widely used in clinical trials, epidemiological research, quality of life (QOL) studies, routine clinical care, and medical surveillance. The Patient Reported Outcomes Measurement Information System (PROMIS) is a system of reliable and standardized measures of PROs developed with Item Response Theory (IRT) using latent scores. Power estimation is critical to clinical trials and research designs. However, in clinical trials with PROs as endpoints, observed scores are often used to calculate power rather than latent scores. In this paper, we conducted a series of simulations to compare the power obtained with IRT latent scores, including Bayesian IRT, Frequentist IRT, and observed scores, focusing on small sample size common in pilot studies and Phase I/II trials. Taking the PROMIS depression measures as an example, we simulated data and estimated power for two-armed clinical trials manipulating the following factors: sample size, effect size, and number of items. We also examined how misspecification of effect size affected power estimation. Our results showed that the Bayesian IRT, which incorporated prior information into latent score estimation, yielded the highest power, especially when sample size was small. The effect of misspecification diminished as sample size increased. For power estimation in two-armed clinical trials with standardized PRO endpoints, if a medium effect size or larger is expected, we recommend BIRT simulation with well-grounded informative priors and a total sample size of at least 40.

From Seeking to Sharing: Pathways of Environmental Risk Information on Social Media and the Roles of Outcome Expectations, Efficacy, and AI-Generated Content

This study explores the motivations driving Chinese consumers to share information about aquatic product safety risks related to radioactive wastewater discharge on social media, with a specific focus on the role of Artificial Intelligence Generated Content (AIGC) and prior information-seeking activities. The study examines key variables such as risk perception, outcome expectations, and the influence of AIGC literacy in shaping attitudes, social norms, and sharing efficacy. Data from 501 participants were analyzed using structural equation modeling. The findings indicate that risk evaluations and personal outcome expectations significantly motivate information sharing, while social expectations deter it. Additionally, AIGC literacy enhances both knowledge acquisition and sharing efficacy, while prior information seeking increases self-efficacy and risk perception, further fostering information sharing. These results provide critical insights into how environmental risk information is processed and disseminated on social media, emphasizing AIGC’s role in influencing this process.

Optimization design and application of library face recognition access control system based on improved PCA.

The application of face recognition technology in Library Access Control System (LACS) has an important impact on improving the security and management efficiency of the library. However, the traditional face recognition methods have some limitations in the face of complex environmental conditions such as illumination and posture change. To solve this problem, an improved method combining the Aggregating Spatial Embeddings for Face Recognition (ASEF) algorithm and Principal Component Analysis (PCA) is proposed. The PCA algorithm is optimized by introducing beta prior and full probability Bayesian model. In addition, the research also integrates K-means Clustering Algorithm (KA) to further improve the accuracy and efficiency of face recognition. The experiment showed that the improved PCA method had an average recognition rate of 92.6%, an average recognition speed of 0.40s, and higher accuracy compared to other related methods, reaching 96%. In practical applications, the system quickly and accurately completes the identification of personnel entry and exit, and improves the efficiency and security of library access management.

Meta-Learning Enables Complex Cluster-Specific Few-Shot Binding Affinity Prediction for Protein-Protein Interactions.

Predicting protein-protein interaction (PPI) binding affinities in unseen protein complex clusters is essential for elucidating complex protein interactions and for the targeted screening of peptide- or protein-based drugs. We introduce MCGLPPI++, a meta-learning framework designed to improve the adaptability of pretrained geometric models in such scenarios. To effectively boost the meta-learning optimization by injecting prior intersample distribution knowledge, three specially designed training sample cluster splitting patterns based on protein interaction interfaces are introduced. Additionally, MCGLPPI++ is equipped with an independent energy component which explicitly models interface nonbonded interaction energies closely related to the strengths of PPIs. To validate our approach, we curate a new data set featuring a challenging test cluster of T-cell receptors binding to antigenic peptide-MHC molecules (TCR-pMHC). Experimental results show that geometric models enhanced by the MCGLPPI++ framework achieve significantly more robust binding affinity predictions after fine-tuning on a few samples from this novel cluster compared to their vanilla counterparts, which demonstrates the effectiveness of the framework.

FlowBack: A Generalized Flow-Matching Approach for Biomolecular Backmapping.

Coarse-grained models have become ubiquitous in biomolecular modeling tasks aimed at studying slow dynamical processes such as protein folding and DNA hybridization. These models can considerably accelerate sampling but it remains challenging to accurately and efficiently restore all-atom detail to the coarse-grained trajectory, which can be vital for detailed understanding of molecular mechanisms and calculation of observables contingent on all-atom coordinates. In this work, we introduce FlowBack as a deep generative model employing a flow-matching objective to map samples from a coarse-grained prior distribution to an all-atom data distribution. We construct our prior distribution to be agnostic to the coarse-grained map and molecular type. A protein-specific model trained on ∼65k structures from the Protein Data Bank achieves state-of-the-art performance on structural metrics compared to previous generative and rules-based approaches in applications to static PDB structures, all-atom simulations of fast-folding proteins, and coarse-grained trajectories generated by a machine-learned force field. A DNA-protein model trained on ∼1.5k DNA-protein complexes achieves excellent reconstruction and generative capabilities on static DNA-protein complexes from the Protein Data Bank as well as on out-of-distribution coarse-grained dynamical simulations of DNA-protein complexation. FlowBack offers an accurate, efficient, and easy-to-use tool to recover all-atom structures from coarse-grained molecular simulations with higher robustness and fewer steric clashes than previous approaches. We make FlowBack freely available to the community as an open source Python package.

Electrical resistivity tomography data inversion using prior information for tunnel prospecting: A case study from southwestern China

AbstractThe direct current (DC) resistivity method is extensively used to predict water‐inrush disasters in tunnel prospecting. However, during DC resistivity inversion, different initial models can significantly affect the inversion results, often resulting in convergence at a local optimum. To overcome these challenges, we propose a new method for DC data inversion that uses prior information as a reference model. First, the resistivity distribution of the surrounding rock mass was estimated based on detailed geological analysis. Next, an initial homogeneous resistivity model was constructed by averaging the observed tunnel resistivity values. Finally, the initial model was developed by incorporating borehole rock samples and water content data. The effectiveness of the method' was assessed using a series of synthetic models of typical water‐bearing structures. We then applied this approach to the Laomacao Tunnel in the Yunnan Central Water Diversion Project (southwestern China), where drilling data were used as a priori information to optimize the initial model together with the average tunnel resistivity values, successfully identifying the water‐bearing structure ahead of the tunnel face. Overall, the proposed method enhances the understanding of sudden surges, aiding in the prevention and control of water disasters in tunnels.

One‐Fit‐All Transformer for Multimodal Geophysical Inversion: Method and Application

AbstractThe increasing variety of geophysical data sets enhances inversion constraints but also poses significant challenges for conventional methods in terms of precision and efficiency due to factors like non‐linearity and limited observation coverage. While deep learning has the potential to address these challenges, issues such as data representation and domain transformation limit the development of universally applicable deep learning‐based inversion methods across various geophysical scenarios. To address these issues, we propose a novel Transformer‐based inversion framework named G(eophysics)‐Query for multimodal geophysical data by expanding the vector representation concept and incorporating query adjusting in the attention mechanism. It can adapt to diverse inversion settings—across different observation systems, targets, and prior information—within a single network, and thus eliminates the need for specific design adjustments and repeated training. Additionally, the framework incorporates a novel semi‐supervised training strategy to explore empirical relationships for prediction uncertainty estimation, ultimately providing more comprehensive inversion results. The framework is applied in the joint inversion of surface waves and receiver functions and successfully obtains the lithospheric structure of the conterminous U.S., demonstrating high precision, efficiency, and universal capability.

Attractive and repulsive visual aftereffects depend on stimulus contrast.

Visual perception has been described as a dynamic process where incoming visual information is combined with what has been seen before to form the current percept. Such a process can result in multiple visual aftereffects that can be attractive toward or repulsive away from past visual stimulation. A lot of research has been conducted on what functional role the mechanisms that produce these aftereffects may play. However, there is a lack of understanding of the role of stimulus uncertainty on these aftereffects. In this study, we investigate how the contrast of a stimulus affects the serial aftereffects it induces and how the stimulus itself is affected by these effects depending on its contrast. We presented human observers with a series of Gabor patches and monitored how the perceived orientation of stimuli changed over time with the systematic manipulation of orientation and contrast of presented stimuli. We hypothesized that repulsive serial effects would be stronger for the judgment of high-contrast than low-contrast stimuli, but the other way around for attractive serial effects. Our experimental findings confirm such a strong interaction between contrast and sign of aftereffects. We present a Bayesian model observer that can explain this interaction based on two principles, the dynamic changes of orientation-tuned channels in short timescales and the slow integration of prior information over long timescales. Our findings have strong implications for our understanding of orientation perception and can inspire further work on the identification of its neural mechanisms.