Prediction Variance Research Articles

Deep learning-based correction for time truncation in cerebral computed tomography perfusion.

Cerebral computed tomography perfusion (CTP) imaging requires complete acquisition of contrast bolus inflow and washout in the brain parenchyma; however, time truncation undoubtedly occurs in clinical practice. To overcome this issue, we proposed a three-dimensional (two-dimensional + time) convolutional neural network (CNN)-based approach to predict missing CTP image frames at the end of the series from earlier acquired image frames. Moreover, we evaluated three strategies for predicting multiple time points. Seventy-two CTP scans with 89 frames and eight slices from a publicly available dataset were used to train and test the CNN models capable of predicting the last 10 image frames. The prediction strategies were single-shot prediction, recursive multi-step prediction, and direct-recursive hybrid prediction.Single-shot prediction predicted all frames simultaneously, while recursive multi-step prediction used prior predictions as input for subsequent steps, and direct-recursive hybrid prediction employed separate models for each step with prior predictions as input for the next step. The accuracies of the predicted image frames were evaluated in terms of image quality, bolus shape, and clinical perfusion parameters. We found that the image quality metrics were superior when multiple CTP images were predicted simultaneously rather than recursively. The bolus shape also showed the highest correlation (r = 0.990, p < 0.001) and the lowest variance (95% confidence interval, -453.26-445.53) in the single-shot prediction. For all perfusion parameters, the single-shot prediction had the smallest absolute differences from ground truth. Our proposed approach can potentially minimize time truncation errors and support the accurate quantification of ischemic stroke.

Efficient augmented radial basis function for reliability analysis with dynamic importance sampling

Conducting reliability analysis on complex systems, particularly those characterized by minimal failure probabilities and computationally intensive models, presents significant challenges. This article introduces a reliability analysis method combining the adaptive augmented radial basis function and the dynamic importance sampling (ARBF-DIS) algorithm. This method improves the generalization capacity of the augmented radial basis function at both local and global scales through the application of an anisotropic technique and orthogonal conditions. By using efficient cross-validation, the methodology facilitates the training of the metamodel, enabling it to provide predictive variance essential for constructing an adaptive sampling learning function. A key aspect of this approach is its capability to address the issue of low failure probabilities through dynamic importance sampling, which involves calculating an optimal importance sampling function derived from the current metamodel. The effectiveness and precision of the ARBF-DIS method are demonstrated through two analytical examples and a solid rocket motor reliability problem.

Energy loss and efficiency investigations of a dual clutch transmission based on adaptive experiment and analytical simulation

Energy transfer improvement and energy consumption reduction are getting more crucial in the automotive sector. As one of the essential elements directly influencing energy transfer in the powertrain system, transmission losses have been widely analyzed with both simulation and experimental approaches. For a better characterization of individual losses, this paper investigates and compares most of the available analytical models for further system implementation. Regarding laboratory experiments, an adaptive sampling strategy is presented and analyzed for various operating conditions. Compared to the traditional factorial design, the applied Gaussian Process Regression (GPR) delivers both estimations and their predictive variances in the entire experiment process. Besides, new measurement candidates can be determined with the prior information from GPR model. The analytical simulation and adaptive measurement are employed to investigate a seven-speed transmission. With both approaches, a detailed discussion about total transmission losses and individual loss mechanisms are performed under various operating conditions.

Sequential metamodel‐based approaches to level‐set estimation under heteroscedasticity

AbstractThis paper proposes two sequential metamodel‐based methods for level‐set estimation (LSE) that leverage the uniform bound built on stochastic kriging: predictive variance reduction (PVR) and expected classification improvement (ECI). We show that PVR and ECI possess desirable theoretical performance guarantees and provide closed‐form expressions for their respective sequential sampling criteria to seek the next design point for performing simulation runs, allowing computationally efficient one‐iteration look‐ahead updates. To enhance understanding, we reveal the connection between PVR and ECI's sequential sampling criteria. Additionally, we propose integrating a budget allocation feature with PVR and ECI, which improves computational efficiency and potentially enhances robustness to the impacts of heteroscedasticity. Numerical studies demonstrate the superior performance of the proposed methods compared to state‐of‐the‐art benchmarking approaches when given a fixed simulation budget, highlighting their effectiveness in addressing LSE problems.

The role of cortical structural variance in deep learning-based prediction of fetal brain age.

Deep-learning-based brain age estimation using magnetic resonance imaging data has been proposed to identify abnormalities in brain development and the risk of adverse developmental outcomes in the fetal brain. Although saliency and attention activation maps have been used to understand the contribution of different brain regions in determining brain age, there has been no attempt to explain the influence of shape-related cortical structural features on the variance of predicted fetal brain age. We examined the association between the predicted brain age difference (PAD: predicted brain age-chronological age) from our convolution neural networks-based model and global and regional cortical structural measures, such as cortical volume, surface area, curvature, gyrification index, and folding depth, using regression analysis. Our results showed that global brain volume and surface area were positively correlated with PAD. Additionally, higher cortical surface curvature and folding depth led to a significant increase in PAD in specific regions, including the perisylvian areas, where dramatic agerelated changes in folding structures were observed in the late second trimester. Furthermore, PAD decreased with disorganized sulcal area patterns, suggesting that the interrelated arrangement and areal patterning of the sulcal folds also significantly affected the prediction of fetal brain age. These results allow us to better understand the variance in deep learning-based fetal brain age and provide insight into the mechanism of the fetal brain age prediction model.

Water Resource Utilization Assessment in China Based on the Dynamic Relationship between Economic Growth and Water Use

Water scarcity has significantly hampered China’s economic, social, and environmental development. Ensuring sustainable water utilization is crucial given the mounting water stress accompanying continuous economic growth. A quantitative water resource forewarning model was constructed using the vector autoregressive (VAR) model. By analyzing the key indicators related to water systems and GDP data from 2001 to 2022, the VAR model revealed the long-term dynamic correlation between water consumption and economic growth using generalized impulse response, co-integration, and predictive variance decomposition analyses. The results revealed the presence of a long-term equilibrium between water consumption and economic growth, with a stable co-integration relationship and an optimal lag period of one year. The positive impact of water consumption on economic development increased during the 2001–2022 period, indicating a rising dependence of GDP on water resources. Water usage rose with economic development, while the water resource carrying capacity remained high and continued to grow. Based on the generalized impulse response, co-integration, and predictive variance decomposition analyses, this study predicted water-use-related indicators, providing vital early warnings for China’s water environment carrying capacity from 2023 to 2050. This enabled informed decision-making and fostered sustainable water management practices for the future.

A catalogue of asteroseismically calibrated ages for APOGEE DR17

Context. The formation history and evolution of the Milky Way through cosmological time is a complex field of research requiring the sampling of highly accurate stellar ages for all Galaxy components. Such highly reliable ages are starting to become available thanks to the synergy of asteroseismology, spectroscopy, stellar modelling, and machine learning analysis in the era of all-sky astronomical surveys. Aims. Our goal is to provide an accurate list of ages for the Main Red Star Sample of the APOGEE DR17 catalogue. In order to reach this goal, ages obtained under asteroseismic constraints are used to train a machine learning model. Methods. As our main objective is to obtain reliable age predictions without the need for asteroseismic parameters, the optimal choice of stellar non-asteroseismic parameters was investigated to obtain the best performances on the test set. The stellar parameters Teff and L, the abundances of [CI/N],[Mg/Ce], and [α/Fe], the U(LSR) velocity, and the vertical height from the Galactic plane ‘Z’ were used to predict ages with a categorical gradient boost decision trees model. The model was trained on two merged samples of the TESS Southern Continuous Viewing Zone and the Second APOKASC catalogue to avoid a data shift and to improve the reliability of the predictions. Finally, the model was tested on an independent data set of the K2 Galactic Archaeology Program. Results. A model with a median fractional age error of 20.8% is obtained. Its prediction variance between the validation and the training set is 4.77%. For stars older than 3 Gyr, the median fractional error in age ranges from 7% to 23%. For stars with ages ranging from 1 to 3 Gyr, the median fractional error in age ranges from 26% to 28%. For stars younger than 1 Gyr, the median fractional error is 43%. The optimised model applies to 125 445 stars from the Main Red Star Sample of the APOGEE DR17 catalogue. Our analysis of the ages confirms previous findings regarding the flaring of the young Galactic disc towards its outer regions. Additionally, we find an age gradient among the youngest stars within the Galactic plane. Finally, we identify two groups of a few metal-poor ([Fe/H] &lt; −1 dex) young stars (Age &lt; 2 Gyr) with similar peculiar chemical abundances and halo kinematics. These are likely the outcomes of the predicted third and latest episode of gas infall in the solar vicinity (~2.7 Gyr ago). Conclusions. We make a catalogue of asteroseismically calibrated ages for 125 445 red giants from the APOGEE DR17 catalogue available to the community. The analysis of the associated stellar parameters corroborates the predictions of different literature models.

A comparison of far-field explosive loads by a selection of current and emerging blast software

In blast resistant design, simulation data may be used to generate the explosive loads that would be witnessed by the detonation of a high explosive device. There are many software packages available to simulate explosions, and this study aims to provide a comparison of a selection of them, including some recently available, with the aim of forming an understanding of the potential accuracies and speed of these in far-field explosive prediction. Software selection criteria were formed by a literature survey to highlight the commonly commercially used programmes. Each software package was used to predict the far-field effects of overpressure and specific impulse from two explosive charges against a reflective target. The data generated by these simulations were then critically compared and reviewed, both with respect to each other and, for the 100 kg charge size, with respect to experimental data. It shows that for the simulated scenarios, after domain and cell size fidelity studies have been performed, the choice of software may lead to a variance in prediction in the peak overpressure of up to 50%, specific impulse of up to 15%, and simulation run times by a factor of 600.

Few-Shot Classification of Wafer Bin Maps Using Transfer Learning and Ensemble Learning

Abstract The high cost of collecting and annotating wafer bin maps (WBMs) necessitates few-shot WBM classification, i.e., classifying WBM defect patterns using a limited number of WBMs. Existing few-shot WBM classification algorithms mainly utilize meta-learning methods that leverage knowledge learned in several episodes. However, meta-learning methods require a large amount of additional real WBMs, which can be unrealistic. To help train a network with a few real-WBMs while avoiding this challenge, we propose the use of simulated WBMs to pre-train a classification model. Specifically, we employ transfer learning by pre-training a classification network with sufficient amounts of simulated WBMs and then fine-tuning it with a few real-WBMs. We further employ ensemble learning to overcome the overfitting problem in transfer learning by fine-tuning multiple sets of classification layers of the network. A series of experiments on a real-dataset demonstrate that our model outperforms the meta-learning methods that are widely used in few-shot WBM classification. Additionally, we empirically verify that transfer and ensemble learning, the two most important yet simple components of our model, reduce the prediction bias and variance in few-shot scenarios without a significant increase in training time.

Depression Analysis Using Face Recognition System

Stress is a part of life it is an unpleasant state of emotional arousal that people experience in situations like working for long hours in front of a computer. Computers have become a way of life, much life is spent on computers and hence we are therefore more affected by the ups and downs that they cause us. One cannot just completely avoid their work on computers but one can at least control his/her usage when being alarmed about him being stressed at a certain point of time. Monitoring the emotional status of a person who is working in front of a computer for a longer duration is crucial for the safety of a person. In this work, a real-time non-intrusive videos are captured, which detects the emotional status of a person by analyzing the facial expression. In each frame of the video, we identify a distinct emotion and assess stress levels over consecutive hours of captured footage. Our approach involves employing a methodology enabling model training and analysis of predictive variances. Utilizing Theano, a Python framework, we enhance both execution and development times for the linear regression model, serving as the deep learning algorithm in this context. Experimental findings indicate that our system effectively performs across diverse age groups with a generic model.

Predicting the Time-Dependent Toxicities of Binary Mixtures of Five Antibiotics to Vibrio qinghaiensis sp.-Q67 Based on the QSAR Model.

Antibiotics may be exposed in a mixed state in natural environments. The toxicity of antibiotic mixtures exhibits time-dependent characteristics, and data on the time-dependent toxicity of antibiotic mixtures is also relatively lacking. In this study, the toxicities of 45 binary mixtures composed of five antibiotics were investigated against Vibrio qinghaiensis sp.-Q67 (Q67) at multiple exposure times (4, 6, 8, 10, and 12 h). Quantitative structure-activity relationship (QSAR) models were developed for predicting the time-dependent toxicities of 45 binary mixtures. The results showed that the best QSAR models presented coefficient of determination (R 2) of (0.818-0.913) and explained variance in prediction leave-one-out (Q 2 LOO) of (0.781-0.894) and predictive ability (Q 2 F1, Q 2 F2, Q 2 F3 > 0.682, concordance correlation coefficient > 0.859). The R 2 values of QSAR models outperformed the R 2 (0.628-0.810) of the conventional concentration addition models and the R 2 (0.654-0.792) of the independent action models. Furthermore, the QSAR models showed higher R 2 and Q 2 LOO values at 4 h compared to other exposure times. Specifically, the model at the 30% effective concentration (EC30) had R 2 of 0.902 and Q 2 LOO of 0.883, while the model at the 50% effective concentration (EC50) had R 2 of 0.913 and Q 2 LOO of 0.894. The CATS2D_04_DP descriptor was found to be the most dominant and negatively correlated factor influencing the toxicity of mixed antibiotics against Q67 in the nine QSAR models developed over five exposure times. The reduction in the number of DP pharmacophore point pairs with a topological distance of 4 in the represented molecules is the primary cause for the rise in the time-dependent toxicity of the antibiotics against Q67.

Automatic landslide detection and visualization by using deep ensemble learning method

Rapid detection of damages occurring as a result of natural disasters is vital for emergency response. In recent years, remote sensing techniques have been commonly used for the automatic categorization and localization of such events using satellite images. Trained based on natural disaster images, a convolutional neural network (CNN) has been applied as a highly successful method, with its ability to reveal outstanding features. Studies aiming to detect target points obtained as a result of extracting visual features from natural images within these networks have achieved their goals. In this study, ensemble learning methods have been suggested as a means to develop the detection of landslide areas from landslide satellite images. Landslide image dataset has been trained for their categorization in CNN models and then they have been used again to localize landslide regions. While model predictions develop overall performance and status, different ensemble strategies have been used and integrated to reduce the sensitivity to prediction variance and training data. Class-selective relevance mapping (CRM) has been used to visualize individual CNN models and ensemble learned behaviors. As a result of the comparisons made based on mean average precision metrics and the criteria of intersection over union, model ensembles have proved to show higher localization performance than any other individual model.

Arbitrariness and Social Prediction: The Confounding Role of Variance in Fair Classification

Variance in predictions across different trained models is a significant, under-explored source of error in fair binary classification. In practice, the variance on some data examples is so large that decisions can be effectively arbitrary. To investigate this problem, we take an experimental approach and make four overarching contributions. We: 1) Define a metric called self-consistency, derived from variance, which we use as a proxy for measuring and reducing arbitrariness; 2) Develop an ensembling algorithm that abstains from classification when a prediction would be arbitrary; 3) Conduct the largest to-date empirical study of the role of variance (vis-a-vis self-consistency and arbitrariness) in fair binary classification; and, 4) Release a toolkit that makes the US Home Mortgage Disclosure Act (HMDA) datasets easily usable for future research. Altogether, our experiments reveal shocking insights about the reliability of conclusions on benchmark datasets. Most fair binary classification benchmarks are close-to-fair when taking into account the amount of arbitrariness present in predictions -- before we even try to apply any fairness interventions. This finding calls into question the practical utility of common algorithmic fairness methods, and in turn suggests that we should reconsider how we choose to measure fairness in binary classification.

A Temperature Prediction Model for Flexible Electronic Devices Based on GA-BP Neural Network and Experimental Verification.

The problem that the thermal safety of flexible electronic devices is difficult to evaluate in real time is addressed in this study by establishing a BP neural network (GA-BPNN) temperature prediction model based on genetic algorithm optimisation. The model uses a BP neural network to fit the functional relationship between the input condition and the steady-state temperature of the equipment and uses a genetic algorithm to optimise the parameter initialisation problem of the BP neural network. To overcome the challenge of the high cost of obtaining experimental data, finite element analysis software is used to simulate the temperature results of the equipment under different working conditions. The prediction variance of the GA-BPNN model does not exceed 0.57 °C and has good robustness, as the model is trained according to the simulation data. The study conducted thermal validation experiments on the temperature prediction model for this flexible electronic device. The device reached steady state after 1200 s of operation at rated power. The error between the predicted and experimental results was less than 0.9 °C, verifying the validity of the model's predictions. Compared with traditional thermal simulation and experimental methods, this model can quickly predict the temperature with a certain accuracy and has outstanding advantages in computational efficiency and integrated application of hardware and software.

Learning manifolds from non-stationary streams

Streaming adaptations of manifold learning based dimensionality reduction methods, such as Isomap, are based on the assumption that a small initial batch of observations is enough for exact learning of the manifold, while remaining streaming data instances can be cheaply mapped to this manifold. However, there are no theoretical results to show that this core assumption is valid. Moreover, such methods typically assume that the underlying data distribution is stationary and are not equipped to detect, or handle, sudden changes or gradual drifts in the distribution that may occur when the data is streaming. We present theoretical results to show that the quality of a manifold asymptotically converges as the size of data increases. We then show that a Gaussian Process Regression (GPR) model, that uses a manifold-specific kernel function and is trained on an initial batch of sufficient size, can closely approximate the state-of-art streaming Isomap algorithms, and the predictive variance obtained from the GPR prediction can be employed as an effective detector of changes in the underlying data distribution. Results on several synthetic and real data sets show that the resulting algorithm can effectively learn lower dimensional representation of high dimensional data in a streaming setting, while identifying shifts in the generative distribution. For instance, key findings on a Gas sensor array data set show that our method can detect changes in the underlying data stream, triggered due to real-world factors, such as introduction of a new gas in the system, while efficiently mapping data on a low-dimensional manifold.

Relationship between prediction accuracy and uncertainty in compound potency prediction using deep neural networks and control models

The assessment of prediction variance or uncertainty contributes to the evaluation of machine learning models. In molecular machine learning, uncertainty quantification is an evolving area of research where currently no standard approaches or general guidelines are available. We have carried out a detailed analysis of deep neural network variants and simple control models for compound potency prediction to study relationships between prediction accuracy and uncertainty. For comparably accurate predictions obtained with models of different complexity, highly variable prediction uncertainties were detected using different metrics. Furthermore, a strong dependence of prediction characteristics and uncertainties on potency levels of test compounds was observed, often leading to over- or under-confident model decisions with respect to the expected variance of predictions. Moreover, neural network models responded very differently to training set modifications. Taken together, our findings indicate that there is only little, if any correlation between compound potency prediction accuracy and uncertainty, especially for deep neural network models, when predictions are assessed on the basis of currently used metrics for uncertainty quantification.

Structural optimization of new media communication culture industry on the impact of digital innovation on marine economic growth

This study investigates the impact of digital innovation on the growth of the marine economy by optimizing the structure of the new media communication culture industry. Through the simulation of the particle swarm optimization (PSO) and Back Propagation (BP) computational network model, the Gross Domestic Production (GDP) of the marine economic zone is predicted, and the prediction effect of the PSO-BP computational network model on multivariate spatiotemporal transitions is further verified. The results show that the PSO-BP model has less prediction variance and a better fit than the BP model. The analysis of the joint prediction model reveals that there are differences in the publicity effects of different self-media modes in various types of sea areas. In terms of the economic peak of the overall data, Bilibili has the best publicity effect, followed by WeChat, Weibo, and Shake. The results show that the trend of macroeconomic indicators and the size of values in different regions affect the prediction effect. Region A has a higher annual average rate of change of macroeconomic indicators and has a better prediction effect. Region B has a faster macroeconomic development and has a more fluctuating prediction effect. Regions C and D have their advantages and disadvantages of the prediction effect under the self-media model.

Gaussian Process Emulation for High-Dimensional Coupled Systems

Complex coupled multiphysics simulations are ubiquitous in science and engineering. Evaluating these numerical simulators is often costly which limits our ability to run them sufficiently often for forward uncertainty quantification. Furthermore outputs are generally not scalar quantities but depend on space and/or time. Gaussian process emulators are statistical surrogates which can approximate the output of the complex computer models at untested inputs quickly while also providing uncertainty information about the accuracy of evaluating the emulator rather than the full physical model. GP emulators were originally developed in the context of scalar output from a single physical model but have since been extended to vector-valued quantities of interest (parallel partial emulators) and to coupled physics by connecting two independent emulators, one for each type of physics (linked emulation). The parallel partial linked GP emulator (PPLE) developed in this work combines the efficiency of a shared correlation structure with the accuracy of linked emulators to produce a new tool for emulating compositions of functions with vector-valued output. The PPLE applied to two numerical experiments out-performs direct emulation of the output composite function producing results with smaller average prediction error and variance.

Optimizing machine learning yield functions using query-by-committee for support vector classification with a dynamic stopping criterion

In the field of materials engineering, the accurate prediction of material behavior under various loading conditions is crucial. Machine Learning (ML) methods have emerged as promising tools for generating constitutive models straight from data, capable of describing complex material behavior in a more flexible way than classical constitutive models. Yield functions, which serve as foundation of constitutive models for plasticity, can be properly described in a data-oriented manner using ML methods. However, the quality of these descriptions heavily relies on the availability of sufficient high-quality and representative training data that needs to be generated by fundamental numerical simulations, experiments, or a combination of both. The present paper addresses the issue of data selection, by introducing an active learning approach for Support Vector Classification (SVC) and its application in training an ML yield function with suitable data. In this regard, the Query-By-Committee (QBC) algorithm was employed, guiding the selection of new training data points in regions of the feature space where a committee of models shows significant disagreement. This approach resulted in a marked reduction in the variance of model predictions throughout the active learning process. It was also shown that the rate of decrease in the variance went along with an increase in the quality of the trained model, quantified by the Matthews Correlation Coefficient (MCC). This demonstrated the effectiveness of the approach and offered us the possibility to define a dynamic stopping criterion based on the variance in the committee results.

RegionalMRI volumetry using NeuroQuant versus visual rating scales in patients with cognitive impairment and dementia.

The aims were to compare the novel regional brain volumetric measures derived by the automatic software NeuroQuant (NQ) with clinically used visual rating scales of medial temporal lobe atrophy (MTA), global cortical atrophy-frontal (GCA-f), and posterior atrophy (PA) brain regions, assessing their diagnostic validity, and to explore if combining automatic and visual methods would increase diagnostic prediction accuracy. Brain magnetic resonance imaging (MRI) examinations from 86 patients with subjective and mild cognitive impairment (i.e., non-dementia, n = 41) and dementia (n = 45) from the Memory Clinic at Oslo University Hospital were assessed using NQ volumetry and with visual rating scales. Correlations, receiver operating characteristic analyses calculating area under the curves (AUCs) for diagnostic accuracy, and logistic regression analyses were performed. The correlations between NQ volumetrics and visual ratings of corresponding regions were generally high between NQ hippocampi/temporal volumes and MTA (r = -0.72/-0.65) and between NQ frontal volume and GCA-f (r = -0.62) but lower between NQ parietal/occipital volumes and PA (r = -0.49/-0.37). AUCs of each region, separating non-dementia from dementia, were generally comparable between the two methods, except that NQ hippocampi volume did substantially better than visual MTA (AUC = 0.80vs. 0.69). Combining both MRI methods increased only the explained variance of the diagnostic prediction substantially regarding the posterior brain region. The findings of this study encourage the use of regional automatic volumetry in locations lacking neuroradiologists with experience in the rating of atrophy typical of neurodegenerative diseases, and in primary care settings.