Uncertainty Estimation Approach Research Articles

Exploring and Exploiting Multi-modality Uncertainty for Tumor Segmentation on PET/CT.

Despite the success of deep learning methods in multi-modality segmentation tasks, they typically produce a deterministic output, neglecting the underlying uncertainty. The absence of uncertainty could lead to over-confident predictions with catastrophic consequences, particularly in safety-critical clinical applications. Recently, uncertainty estimation has attracted increasing attention, offering a measure of confidence associated with machine decisions. Nonetheless, existing uncertainty estimation approaches primarily focus on single-modality networks, leaving the uncertainty of multi-modality networks a largely under-explored domain. In this study, we present the first exploration of multi-modality uncertainties in the context of tumor segmentation on PET/CT. Concretely, we assessed four well-established uncertainty estimation approaches across various dimensions, including segmentation performance, uncertainty quality, comparison to single-modality uncertainties, and correlation to the contradictory information between modalities. Through qualitative and quantitative analyses, we gained valuable insights into what benefits multi-modality uncertainties derive, what information multi-modality uncertainties capture, and how multi-modality uncertainties correlate to information from single modalities. Drawing from these insights, we introduced a novel uncertainty-driven loss, which incentivized the network to effectively utilize the complementary information between modalities. The proposed approach outperformed the backbone network by 4.53 and 2.92 Dices in percentages on two PET/CT datasets while achieving lower uncertainties. This study not only advanced the comprehension of multi-modality uncertainties but also revealed the potential benefit of incorporating them into the segmentation network. The code is available at https://github.com/HUST-Tan/MMUE.

Monte Carlo analysis of masonry structures under tsunami action: Reliability of lognormal fragility curves and overall uncertainty prediction

Tsunami vulnerability of coastal buildings has gained more and more interest in recent years, in the consciousness of what losses may be caused. The improvement of the available approaches for the quantitative estimation of the probability of building damage and for defining possible strategies for risk mitigation is an actual goal. In this framework, several authors have provided empirical fragility curves based on field surveys after tsunamis. Nevertheless, a predictive approach based on analytical fragility curves, which can be extended to many classes of buildings, is essential for the scopes of civil protection and risk mitigation. In this paper, an approach for the construction of fragility curves, proposed for masonry structures under tsunami waves, is discussed and refined in the part regarding the assignment of the uncertainties. Further, an assessment of the reliability of the lognormal fragility distribution is carried out based on a Monte Carlo simulation applied to 4 classes of buildings. Here, it is shown that Monte Carlo analysis allows a direct evaluation of the uncertainties without the need to resort to ambiguous regression analyses and rules of combination of the uncertainties of demand and capacity based on the regression analysis results or other uncertainty estimation approaches.

DUDES: Deep Uncertainty Distillation using Ensembles for Semantic Segmentation

The intersection of deep learning and photogrammetry unveils a critical need for balancing the power of deep neural networks with interpretability and trustworthiness, especially for safety-critical application like autonomous driving, medical imaging, or machine vision tasks with high demands on reliability. Quantifying the predictive uncertainty is a promising endeavour to open up the use of deep neural networks for such applications. Unfortunately, most current available methods are computationally expensive. In this work, we present a novel approach for efficient and reliable uncertainty estimation for semantic segmentation, which we call Deep Uncertainty Distillation using Ensembles for Segmentation (DUDES). DUDES applies student-teacher distillation with a Deep Ensemble to accurately approximate predictive uncertainties with a single forward pass while maintaining simplicity and adaptability. Experimentally, DUDES accurately captures predictive uncertainties without sacrificing performance on the segmentation task and indicates impressive capabilities of highlighting wrongly classified pixels and out-of-domain samples through high uncertainties on the Cityscapes and Pascal VOC 2012 dataset. With DUDES, we manage to simultaneously simplify and outperform previous work on Deep-Ensemble-based Uncertainty Distillation.

Coprecipitation as a One-Step Se Separation for Determination of Isotope Ratios Completed with Revised Uncertainty Evaluation.

This study introduces a simplified purification method for analyzing 82Se/78Se isotope ratios in diverse natural samples using hydride generation MC-ICP-MS. Unlike the thiol resin method, which is time-consuming and sensitive to the concentrations of reagents used at individual stages, our proposed alternative is quicker, simpler, and robust. The procedure involves coprecipitation of selenium with iron hydroxide and dissolution in hydrochloric acid. Combining hydride generation and a second cleanup stage achieves sufficient purification for Se isotope ratio measurements. The method is efficient, taking 3-4 h after sample decomposition, utilizing common reagents [HCl, Fe(NO3)3, NH4Cl] without evaporation or clean lab conditions. Results on 82Se/78Se isotope ratios in various matrices are presented, comparing them with literature data. All isotopic results have been subjected to a newly proposed state-of-the-art approach to uncertainty estimation dedicated to isotope ratio measurements. The approach is based on applying Monte Carlo simulations with consideration of different samples' results normalized by the expected value. By doing that, we obtained estimated uncertainty for any Se sample with the influence of particular measurements on the final estimation included. We employ a Monte Carlo simulation-based uncertainty estimation approach for isotope ratio measurements, providing estimated uncertainty for each selenium sample.

Qualitative and quantitative enhancement of parameter estimation for model-based diagnostics using automatic differentiation with an application to inertial fusion

Parameter estimation using observables is a fundamental concept in the experimental sciences. Mathematical models that represent the physical processes can enable reconstructions of the experimental observables and greatly assist in parameter estimation by turning it into an optimization problem which can be solved by gradient-free or gradient-based methods. In this work, the recent rise in flexible frameworks for developing differentiable scientific computing programs is leveraged in order to dramatically accelerate data analysis of a common experimental diagnostic relevant to laser–plasma and inertial fusion experiments, Thomson scattering. A differentiable Thomson-scattering data analysis tool is developed that uses reverse-mode automatic differentiation (AD) to calculate gradients. By switching from finite differencing to reverse-mode AD, three distinct outcomes are achieved. First, gradient descent is accelerated dramatically to the extent that it enables near real-time usage in laser–plasma experiments. Second, qualitatively novel quantities which require O(103) parameters can now be included in the analysis of data which enables unprecedented measurements of small-scale laser–plasma phenomena. Third, uncertainty estimation approaches that leverage the value of the Hessian become accurate and efficient because reverse-mode AD can be used for calculating the Hessian.

Estimation of airship states and model uncertainties using nonlinear estimators

This Airships are lighter than air vehicles and due to their growing number of applications, they are becoming attractive for the research community. Most of the applications require an airship autonomous flight controller which needs an accurate model and state information. Usually, airship states are affected by noise and states information can be lost in the case of sensor's faults, while airship model is affected by model inaccuracies and model uncertainties. This paper presents the application of nonlinear and Bayesian estimators for estimating the states and model uncertainties of neutrally buoyant airship. It is considered that minimum sensor measurements are available, and data is corrupted with process and measurement noise. A novel lumped model uncertainty estimation approach is formulated where airship model is augmented with six extra state variables capturing the model uncertainty of the airship. The designed estimator estimates the airship model uncertainty along with its states. Nonlinear estimators, Extended Kalman Filter and Unscented Kalman Filter are designed for estimating airship attitude, linear velocities, angular velocities and model uncertainties. While Particle filter is designed for the estimation of airship attitude, linear velocities and angular velocities. Simulations have been performed using nonlinear 6-DOF simulation model of experimental airship for assessing the estimator performances. 1−𝜎 uncertainty bound and error analysis have been performed for the validation. A comparative study of the estimator's performances is also carried out.

Bridging AI and Clinical Practice: Integrating Automated Sleep Scoring Algorithm with Uncertainty-Guided Physician Review.

This study aims to enhance the clinical use of automated sleep-scoring algorithms by incorporating an uncertainty estimation approach to efficiently assist clinicians in the manual review of predicted hypnograms, a necessity due to the notable inter-scorer variability inherent in polysomnography (PSG) databases. Our efforts target the extent of review required to achieve predefined agreement levels, examining both in-domain (ID) and out-of-domain (OOD) data, and considering subjects' diagnoses. A total of 19,578 PSGs from 13 open-access databases were used to train U-Sleep, a state-of-the-art sleep-scoring algorithm. We leveraged a comprehensive clinical database of an additional 8832 PSGs, covering a full spectrum of ages (0-91 years) and sleep-disorders, to refine the U-Sleep, and to evaluate different uncertainty-quantification approaches, including our novel confidence network. The ID data consisted of PSGs scored by over 50 physicians, and the two OOD sets comprised recordings each scored by a unique senior physician. U-Sleep demonstrated robust performance, with Cohen's kappa (K) at 76.2% on ID and 73.8-78.8% on OOD data. The confidence network excelled at identifying uncertain predictions, achieving AUROC scores of 85.7% on ID and 82.5-85.6% on OOD data. Independently of sleep-disorder status, statistical evaluations revealed significant differences in confidence scores between aligning vs discording predictions, and significant correlations of confidence scores with classification performance metrics. To achieve κ ≥ 90% with physician intervention, examining less than 29.0% of uncertain epochs was required, substantially reducing physicians' workload, and facilitating near-perfect agreement. Inter-scorer variability limits the accuracy of the scoring algorithms to ~80%. By integrating an uncertainty estimation with U-Sleep, we enhance the review of predicted hypnograms, to align with the scoring taste of a responsible physician. Validated across ID and OOD data and various sleep-disorders, our approach offers a strategy to boost automated scoring tools' usability in clinical settings.

Uncertainty Estimation and Out-of-Distribution Detection for Deep Learning-Based Image Reconstruction using the Local Lipschitz.

Accurate image reconstruction is at the heart of diagnostics in medical imaging. Supervised deep learning-based approaches have been investigated for solving inverse problems including image reconstruction. However, these trained models encounter unseen data distributions that are widely shifted from training data during deployment. Therefore, it is essential to assess whether a given input falls within the training data distribution. Current uncertainty estimation approaches focus on providing an uncertainty map to radiologists, rather than assessing the training distribution fit. In this work, we propose a method based on the local Lipschitz metric to distinguish out-of-distribution images from in-distribution with an area under the curve of 99.94% for True Positive Rate versus False Positive Rate. We demonstrate a very strong relationship between the local Lipschitz value and mean absolute error (MAE), supported by a Spearman's rank correlation coefficient of 0.8475, to determine an uncertainty estimation threshold for optimal performance. Through the identification of false positives, we demonstrate the local Lipschitz and MAE relationship can guide data augmentation and reduce uncertainty. Our study was validated using the AUTOMAP architecture for sensor-to-image Magnetic Resonance Imaging (MRI) reconstruction. We demonstrate our approach outperforms baseline techniques of Monte-Carlo dropout and deep ensembles as well as the state-of-the-art Mean Variance Estimation (MVE) network approach. We expand our application scope to MRI denoising and Computed Tomography (CT) sparse-to-full view reconstructions using UNET architectures. We show our approach is applicable to various architectures and applications, especially in medical imaging, where preserving diagnostic accuracy of reconstructed images remains paramount.

Indoor replication of outdoor climbing routes: fidelity analysis of digital manufacturing workflow

AbstractThis study aims to evaluate the advantages and criticalities of applying additive manufacturing to produce climbing holds replicating real rocky surfaces. A sample of a rocky surface has been reproduced with a budget-friendly 3D scanner exploiting structured light and made in additive manufacturing. The methodology is designed to build a high-fidelity replica of the rocky surface using only minor geometry modifications to convert a 2D triangulated surface into a 3D watertight model optimised for additive manufacturing. In addition, the research uses a novel design and uncertainty estimation approach. The proposed methodology proved capable of replicating a rocky sample with sub-millimetre accuracy, which is more realistic than conventional screw-on plastic holds currently used in climbing gyms. The advantages can be addressed in terms of customisation, manufacturing cost and time reduction that could lead to real outdoor climbing experiences in indoor environments by coupling additive manufacturing techniques and reverse engineering (RE). However, operating the scanner in a rocky environment and the considerable size of the climbing routes suggest that further research is needed to extend the proposed methodology to real case studies. Further analysis should focus on selecting the best material and additive manufacturing technology to produce structural components for climbing environments.

Short-term bus travel time prediction for transfer synchronization with intelligent uncertainty handling

This paper presents two novel approaches for uncertainty estimation adapted and extended for the multi-link bus travel time problem. The uncertainty is modeled directly as part of recurrent artificial neural networks, but using two fundamentally different approaches: one based on Deep Quantile Regression and the other on Bayesian neural network. Both approaches use a recurrent neural network to predict multiple time steps into the future, but handle the time-dependent uncertainty estimation differently. We present a novel sampling technique in order to aggregate quantile estimates for link level travel time to yield the multi-link travel time distribution needed for a vehicle to travel from its current position to a specific downstream stop point or transfer site.To motivate the relevance of uncertainty-aware models in the domain, we focus on the connection protection application as a case study: An expert system to determine whether a bus driver should hold and wait for a connecting service, thus ensuring the connection, or break the connection and reduce its own delay. Our results show that the proposed quantile sampling method performs overall best for the 80%, 90% and 95% prediction intervals, both for a 15 min time horizon into the future (t+1), but also for the 30 and 45 min time horizon (t+2 and t+3), with a constant, but very small underestimation of the uncertainty interval (1–4 pp.). However, we also show, that the Bayesian model still can outperform the DQR for specific cases. Lastly, we demonstrate how a simple decision support system can take advantage of our uncertainty-aware travel time models to prioritize the difference in travel time uncertainty for bus holding at strategic points, thus reducing the introduced delay for the connection protection application.

Prediction-Uncertainty-Aware Decision-Making for Autonomous Vehicles

Motion prediction is the fundamental input for decision-making in autonomous vehicles. The current motion prediction solutions are designed with a strong reliance on black box predictions based on neural networks (NNs), which is unacceptable for safety-critical applications. Motion prediction with high uncertainty can cause conflicting decisions and even catastrophic results. To address this issue, an uncertainty estimation approach based on the deep ensemble technique is proposed for motion prediction in this paper. Subsequently, the estimated uncertainty is considered in the decision-making module to improve driving safety. Firstly, a motion prediction model based on long short-term memory (LSTM) is built and the deep ensemble technique is utilized to obtain both epistemic and aleatoric uncertainty of the prediction model. Besides, an uncertainty-aware potential field is developed to process the prediction uncertainty. Furthermore, a decision-making framework is proposed based on the model predictive control algorithm that considers the uncertainty-aware potential field, road boundaries, and multiple constraints of vehicle dynamics. Finally, the public available <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">NGSIM</i> , <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HighD</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INTERACTION</i> datasets are used to evaluate the proposed motion prediction model. More importantly, two traffic scenarios are also extracted from <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">NGSIM</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">INTERACTION</i> datasets to verify the effectiveness of the proposed decision-making method and in particular, its real-time performance is shown by employing a hardware-in-the-loop (HiL) experiment bench.

Determining the static characteristic of a measuring current transformer at a reduced load of the metering unit

The object of the research is a measuring current transformer of the electromagnetic type, which is used as part of the electricity metering unit. The current transformers functioning in the mode of reduced primary current is accompanied by significant errors. The existence of such a regime for a long time due to downtime of production equipment leads to a significant underestimation of electricity. This leads to unjustified financial losses for energy supply companies as electricity in many countries has become more expensive. The static characteristic of the measuring current transformer at a reduced load of the metering unit is described by a linear statistical model. The parameters of the model are estimated on the basis of empirical data using methods of covariance analysis. The adequacy of the model is confirmed by analysis of regression residuals. The obtained statistical model of the static characteristic, unlike the known ones, is characterized by universality, as it describes current transformers with an arbitrary transformation ratio within the studied accuracy class of 0.5S. The uncertainty of the current error is estimated using the prediction intervals for the transformers secondary currents as a function of the primary currents, calculated on the basis of the obtained model. The prediction intervals for sample values of the current error were obtained through the application of the proposed approach for uncertainty estimation of such an error at a reduced load of the metering unit. It was found that the current error can reach −20 %. Taking into account the indicated intervals for measuring current transformers at a reduced load of the metering unit during billing will allow energy supply companies to increase the accuracy of commercial electricity metering.

Downscaling ensemble climate projections to urban scale: Brussels's future climate at 1.5 °C, 2 °C, and 3 °C global warming

There is an increasing need to obtain climate projections for cities using an ensemble approach for uncertainty estimation. Yet, current-day computational resources are too limited to dynamically downscale GCM ensembles to urban scale. Here, a recently developed and validated statistical-dynamical computationally-cheap method is employed to downscale ten EURO-CORDEX climate projections over Brussels (Belgium) covering the period 1971–2100. Results show that, under the Paris agreement, summer mean projected temperature in Brussels will rise by 3.6 °C to 4.1 °C [+ − 0.7 °C] on average. The Urban Heat Island (UHI) intensity does not increase under future global warming with even a slight decrease under heatwave (HW) conditions by 0.1 °C (±0.1 °C). However, the number of HW days is projected to be 30.6% and 158.9% higher for the 2 °C and 3 °C Global Warming Levels (GWL), respectively, as compared to 1.5 °C GWL. The heat stress during HW periods also follows the same trend: compared to a 1.5 °C GWL, the number of extreme heat stress days at 2 °C (3 °C) GWL will increase by 29% (91%) on average inside the city. The results can be used in support of adaptation measures, which should be considered for future resilience of the city of Brussels.

Floating solar panels on reservoirs impact phytoplankton populations: A modelling experiment

Floating solar photovoltaic (FPV) deployments are increasing globally as the switch to renewable energy intensifies, representing a considerable water surface transformation. FPV installations can potentially impact aquatic ecosystem function, either positively or negatively. However, these impacts are poorly resolved given the challenges of collecting empirical data for field or modelling experiments. In particular, there is limited evidence on the response of phytoplankton to changes in water body thermal dynamics and light climate with FPV. Given the importance of understanding phytoplankton biomass and species composition for managing ecosystem services, we use an uncertainty estimation approach to simulate the effect of FPV coverage and array siting location on a UK reservoir. FPV coverage was modified in 10% increments from a baseline with 0% coverage to 100% coverage for three different FPV array siting locations based on reservoir circulation patterns. Results showed that FPV coverage significantly impacted thermal properties, resulting in highly variable impacts on phytoplankton biomass and species composition. The impacts on phytoplankton were often dependent on array siting location as well as surface coverage. Changes to phytoplankton species composition were offset by the decrease in phytoplankton biomass associated with increasing FPV coverage. We identified that similar phytoplankton biomass reductions could be achieved with less FPV coverage by deploying the FPV array on the water body's faster-flowing area than the central or slower flowing areas. The difference in response dependent on siting location could be used to tailor phytoplankton management in water bodies. Simulation of water body-FPV interactions efficiently using an uncertainty approach is an essential tool to rapidly develop understanding and ultimately inform FPV developers and water body managers looking to minimise negative impacts and maximise co-benefits.

Unified fatigue life modelling and uncertainty estimation of Ni-based superalloy family with a supervised machine learning approach

Machine learning (ML) approaches, especially the supervised learning methods, show enormous advantages in fatigue investigation, whereas few works follow the interest in the generalization and uncertainty estimation of these data-drive approaches. Within this content, a supervised ML method based on the support vector regression (SVR) for key features identification of fatigue life is presented for the Ni-based superalloy family. The unified SVR model is effective at predicting lives for the Ni-based superalloy family under wide loading conditions and fatigue regimes compared with the classical fatigue life models. In addition, a model fusion method is employed to estimate the uncertainty and data dependency of the SVR model, with which the training strategy produces an important effect on the accuracy and stability of the predicted results. It is found that the coefficient of the uncertainty achieves the optimum where the training percentage is 70% of the modelling samples. After input variable selection by the pairwise Pearson correlation and key feature reorganization, the dimensionality reduced SVR model remains an acceptable accuracy. Predictably, the total strain range and the test frequency are recognized as the highly correlated variables with the fatigue life investigating from the perspective of datasets. The trained ML model and uncertainty estimation approach provide potential tools for fatigue investigation under complex loading conditions. Going forwards, it would be beneficial to generalization ability and uncertainty estimation of a ML model for unified fatigue life modelling.

The impact of hurricane disturbances on a tropical forest: implementing a palm plant functional type and hurricane disturbance module in ED2-HuDi V1.0

Abstract. Hurricanes commonly disturb and damage tropical forests. Hurricane frequency and intensity are predicted to change under the changing climate. The short-term impacts of hurricane disturbances to tropical forests have been widely studied, but the long-term impacts are rarely investigated. Modeling is critical to investigate the potential response of forests to future disturbances, particularly if the nature of the disturbances is changing with climate. Unfortunately, existing models of forest dynamics are not presently able to account for hurricane disturbances. Therefore, we implement the Hurricane Disturbance in the Ecosystem Demography model (ED2) (ED2-HuDi). The hurricane disturbance includes hurricane-induced immediate mortality and subsequent recovery modules. The parameterizations are based on observations at the Bisley Experimental Watersheds (BEW) in the Luquillo Experimental Forest in Puerto Rico. We add one new plant functional type (PFT) to the model – Palm, as palms cannot be categorized into one of the current existing PFTs and are known to be an abundant component of tropical forests worldwide. The model is calibrated with observations at BEW using the generalized likelihood uncertainty estimation (GLUE) approach. The optimal simulation obtained from GLUE has a mean relative error of −21 %, −12 %, and −15 % for stem density, basal area, and aboveground biomass, respectively. The optimal simulation also agrees well with the observation in terms of PFT composition (+1 %, −8 %, −2 %, and +9 % differences in the percentages of “Early”, “Mid”, “Late”, and “Palm” PFTs, respectively) and size structure of the forest (+0.8 % differences in the percentage of large stems). Lastly, using the optimal parameter set, we study the impact of forest initial condition on the recovery of the forest from a single hurricane disturbance. The results indicate that, compared to a no-hurricane scenario, a single hurricane disturbance has little impact on forest structure (+1 % change in the percentage of large stems) and composition (&lt;1 % change in the percentage of each of the four PFTs) but leads to 5 % higher aboveground biomass after 80 years of succession. The assumption of a less severe hurricane disturbance leads to a 4 % increase in aboveground biomass.

NPBDREG: Uncertainty assessment in diffeomorphic brain MRI registration using a non-parametric Bayesian deep-learning based approach.

Quantification of uncertainty in deep-neural-networks (DNN) based image registration algorithms plays a critical role in the deployment of image registration algorithms for clinical applications such as surgical planning, intraoperative guidance, and longitudinal monitoring of disease progression or treatment efficacy as well as in research-oriented processing pipelines. Currently available approaches for uncertainty estimation in DNN-based image registration algorithms may result in sub-optimal clinical decision making due to potentially inaccurate estimation of the uncertainty of the registration stems for the assumed parametric distribution of the registration latent space. We introduce NPBDREG, a fully non-parametric Bayesian framework for uncertainty estimation in DNN-based deformable image registration by combining an Adam optimizer with stochastic gradient Langevin dynamics (SGLD) to characterize the underlying posterior distribution through posterior sampling. Thus, it has the potential to provide uncertainty estimates that are highly correlated with the presence of out of distribution data. We demonstrated the added-value of NPBDREG, compared to the baseline probabilistic VoxelMorph model (PrVXM), on brain MRI image registration using 390 image pairs from four publicly available databases: MGH10, CMUC12, ISBR18 and LPBA40. The NPBDREG shows a better correlation of the predicted uncertainty with out-of-distribution data (r>0.95 vs. r<0.5) as well as a ~7.3 % improvement in the registration accuracy (Dice score, 0.74 vs. 0.69, p≪0.01), and a ~18 % improvement in registration smoothness (percentage of folds in the deformation field, 0.014 vs. 0.017, p≪0.01). Finally, NPBDREG demonstrated a better generalization capability for data corrupted by a mixed structure noise (Dice score of 0.73 vs. 0.69, p≪0.01) compared to the baseline PrVXM approach.

Predicting binding motifs of complex adsorbates using machine learning with a physics-inspired graph representation.

Computational screening in heterogeneous catalysis relies increasingly on machine learning models for predicting key input parameters due to the high cost of computing these directly using first-principles methods. This becomes especially relevant when considering complex materials spaces such as alloys, or complex reaction mechanisms with adsorbates that may exhibit bi- or higher-dentate adsorption motifs. Here we present a data-efficient approach to the prediction of binding motifs and associated adsorption enthalpies of complex adsorbates at transition metals and their alloys based on a customized Wasserstein Weisfeiler-Lehman graph kernel and Gaussian process regression. The model shows good predictive performance, not only for the elemental transition metals on which it was trained, but also for an alloy based on these transition metals. Furthermore, incorporation of minimal new training data allows for predicting an out-of-domain transition metal. We believe the model may be useful in active learning approaches, for which we present an ensemble uncertainty estimation approach.

Uncertainty quantification of phase velocity surface waves multy-modal inversion using machine learning

The paper is devoted to uncertainty quantification of the inverse problem solution of the multichannel analysis of surface waves method - the inversion of the curves of the phase velocity via frequency dependence. The uncertainty estimation approach is based on the Monte Carlo sampling strategy and a multilayer fully connected artificial neural network to approximate nonlinear dependence of shear wave velocity and layers thickness via values of phase velocity surface waves. Frequency-dependent noise in the data and errors of the inverse operator are projected onto the inverse problem solution. The results of unimodal and multimodal inversion are compared on the example of synthetic data processing. The experimental results show that using of machine learning approaches makes it possible to quickly and accurately estimate the posterior probability density of the reconstructed velocity model parameters.

Uncertainty-aware membranous nephropathy classification: A Monte-Carlo dropout approach to detect how certain is the model

ABSTRACT Membranous nephropathy (MN) is among the most common glomerular diseases that cause nephrotic syndrome in adults. To aid pathologists on performing the MN classification task, we proposed here a pipeline consisted of two steps. Firstly, we assessed four deep-learning-based architectures, namely, ResNet-18, MobileNet, DenseNet, and Wide-ResNet. To achieve more reliable predictions, we adopted and extensively evaluated a Monte-Carlo dropout approach for uncertainty estimation. Using a 10-fold cross-validation setup, all models achieved average F1-scores above 92%, where the highest average value of 93.2% was obtained by using Wide-ResNet. Regarding uncertainty estimation with Wide-ResNet, high uncertainty scores were more associated with erroneous predictions, demonstrating that our approach can assist pathologists in interpreting the predictions with high reliability. We show that uncertainty-based thresholds for decision referral can greatly improve classification performance, increasing the accuracy up to 96%. Finally, we investigated how the uncertainty scores relate to complexity scores defined by pathologists.