Uncertainty Estimation Research Articles

Inferring stellar parameters and their uncertainties from high-resolution spectroscopy using invertible neural networks

New spectroscopic surveys will increase the number of astronomical objects in need of characterisation by more than an order of magnitude. Machine learning tools are required to address this data deluge in a fast and accurate fashion. Most machine learning algorithms cannot directly estimate error, making them unsuitable for reliable science. We aim to train a supervised deep-learning algorithm tailored for high-resolution observational stellar spectra. This algorithm accurately infers precise estimates while providing coherent estimates of uncertainties by leveraging information from both the neural network and the spectra. We trained a conditional invertible neural network (cINN) on observational spectroscopic data obtained from the GIRAFFE spectrograph (HR10 and HR21 setups) within the Gaia-ESO survey. A key feature of cINN is its ability to produce the Bayesian posterior distribution of parameters for each spectrum. By analysing this distribution, we inferred stellar parameters and their corresponding uncertainties. We carried out several tests to investigate how parameters are inferred and errors are estimated. We achieved an accuracy of 28K in $T_ eff $, 0.06 dex in $ g$, 0.03 dex in Fe/H $, and between 0.05 dex and 0.17 dex for the other abundances for high-quality spectra. Accuracy remains stable with low signal-to-noise ratio (between 5 and 25) spectra, with an accuracy of 39K in $T_ eff $, 0.08 dex in $ g$, and 0.05 dex in Fe/H $. The uncertainties obtained are well within the same order of magnitude. The network accurately reproduces astrophysical relationships both on the scale of the Milky Way and within smaller star clusters. We created a table containing the new parameters generated by our cINN. This neural network represents a compelling proposition for future astronomical surveys. These derived uncertainties are coherent and can therefore be reused in future works as Bayesian priors.

Evidential Deep Learning: Enhancing Predictive Uncertainty Estimation for Earth System Science Applications

Abstract Robust quantification of predictive uncertainty is a critical addition needed for machine learning applied to weather and climate problems to improve understanding of what is driving prediction sensitivity. Ensembles of machine learning models provide predictive uncertainty estimates in a conceptually simple way but require multiple models for training and prediction, increasing computational cost and latency. Parametric deep learning can estimate uncertainty with one model by predicting the parameters of a probability distribution but does not account for epistemic uncertainty. Evidential deep learning, a technique that extends parametric deep learning to higher-order distributions, can account for both aleatoric and epistemic uncertainty with one model. This study compares the uncertainty derived from evidential neural networks to that obtained from ensembles. Through applications of classification of winter precipitation type and regression of surface layer fluxes, we show evidential deep learning models attaining predictive accuracy rivaling standard methods, while robustly quantifying both sources of uncertainty. We evaluate the uncertainty in terms of how well the predictions are calibrated and how well the uncertainty correlates with prediction error. Analyses of uncertainty in the context of the inputs reveal sensitivities to underlying meteorological processes, facilitating interpretation of the models. The conceptual simplicity, interpretability, and computational efficiency of evidential neural networks make them highly extensible, offering a promising approach for reliable and practical uncertainty quantification in Earth system science modeling. In order to encourage broader adoption of evidential deep learning, we have developed a new Python package, MILES-GUESS (https://github.com/ai2es/miles-guess), that enables users to train and evaluate both evidential and ensemble deep learning.

River suspended-sand flux computation with uncertainty estimation using water samples and high-resolution ADCP measurements

Abstract. Measuring suspended-sand fluxes in rivers remains a scientific challenge due to their high spatial and temporal variability. To capture the vertical and lateral gradients of concentration in the cross-section, measurements with point samples are performed. However, the uncertainty related to these measurements is rarely evaluated, as few studies of the major sources of error exist. Therefore, the aim of this study is to develop a method to determine the cross-sectional sand flux and estimate its uncertainty. This SDC (for sand discharge computing) method combines suspended-sand concentrations from point samples with ADCP (acoustic Doppler current profiler) high-resolution depth and velocity measurements. The MAP (for multitransect averaged profile) method allows obtaining an average of several ADCP transects on a regular grid, including the unmeasured areas. The suspended-sand concentrations are integrated vertically by fitting a theoretical exponential suspended-sand profile to the data using Bayesian modeling. The lateral integration is based on the water depth as a proxy for the local bed shear stress to evaluate the bed concentration and sediment diffusion along the river cross-section. The estimation of uncertainty combines ISO standards and semi-empirical methods with a Bayesian approach to estimate the uncertainty due to the vertical integration. The new method is applied to data collected in four rivers under various hydro-sedimentary conditions: the Colorado, Rhône, Isère, and Amazon rivers, with computed flux uncertainties ranging between 18 % and 32 %. The relative difference between the suspended-sand flux in 21 cases calculated with the proposed SDC method compared to the International Organization for Standardization (ISO) 4363 standard method ranges between −40 % and +23 %. This method that comes with a flexible, open-source code is the first to propose an applicable uncertainty estimation that could be adapted to other flux computation methods.

GGR Handbook of Rock and Mineral Analysis Chapter 1 (Part 2) Sampling as Part of the Measurement Process

This chapter (Sampling as Part of the Measurement Process) is a contribution to the Geostandards and Geoanalytical Research Handbook of Rock and Mineral Analysis – an online textbook that is a fully revised and updated edition of A Handbook of Silicate Rock Analysis (P. J. Potts, 1987, Blackie, Glasgow).Chapter 1 (Part 2) forms part of Section 1 of the handbook dealing with fundamentals of measurement and instrument design. The geochemical measurement process is considered to begin when the primary sample is taken from the sampling target, rather than when that sample arrives at the laboratory. This integration of sampling within the measurement procedure enables both sampling and chemical analysis to be optimised in order to achieve a measurement procedure that is fit for its intended geochemical purpose. The key metric in judging this fitness for purpose, and hence validating a measurement procedure, is the uncertainty of each measurement value. This measurement uncertainty is explained, together with methods to estimate and express it in a way that includes the contribution from sampling, with a worked example. The resultant more realistic estimates of measurement uncertainty are shown to improve the reliability of the geochemical interpretation of measurement values.

A Likelihood-Based Triangulation Method for Uncertainties in Through-Water Depth Mapping

Coastal environments, which are crucial for economic and strategic reasons, heavily rely on accurate bathymetry for safe navigation and resource monitoring. Recent advancements in through-water photogrammetry have shown promise in mapping shallow waters efficiently. However, robust uncertainty modeling methods for these techniques, especially in challenging coastal environments, are lacking. This study introduces a novel likelihood-based approach for through-water photogrammetry, focusing on uncertainties associated with camera pose—a key factor affecting depth mapping accuracy. Our methodology incorporates probabilistic modeling and stereo-photogrammetric triangulation to provide realistic estimates of uncertainty in Water Column Depth (WCD) and Water–Air Interface (WAI) height. Using simulated scenarios for both drone and airborne surveys, we demonstrate that viewing geometry and camera pose quality significantly influence resulting uncertainties, often overshadowing the impact of depth itself. Our results reveal the superior performance of the likelihood ratio statistic in scenarios involving high attitude noise, high flight altitude, and complex viewing geometries. Notably, drone-based applications show particular promise, achieving decimeter-level WCD precision and WAI height estimations comparable to high-quality GNSS measurements when using large samples. These findings highlight the potential of drone-based surveys in producing more accurate bathymetric charts for shallow coastal waters. This research contributes to the refinement of uncertainty quantification in bathymetric charting and sets a foundation for future advancements in through-water surveying methodologies.

The Use of Image‐Based Data and Abundance Modelling Approaches for Predicting the Location of Vulnerable Marine Ecosystems in the South Pacific Ocean

ABSTRACTVulnerable marine ecosystems (VMEs) are typically fragile and slow to recover, thereby making them susceptible to disturbance, including fishing. In the high seas, the United Nations General Assembly (UNGA) requested regional fishery management organisations (RFMOs) to implement measures to prevent significant adverse impacts on VMEs. Here, we predict spatial abundances of 15 taxa, 13 VME indicator taxa, in the South Pacific RFMO (SPRFMO) area. Models used seafloor imagery data, an important advance on previously developed presence‐only predictions, to provide information on spatial variation in taxa abundance that is crucial for better inferring likely location of VMEs, rather than just distribution of VME indicator taxa. Abundance models varied in predictive power (mean R2 ranged 0.02–0.40). Uncertainty estimates of model predictions were developed to inform future spatial planning processes for conservation and management of VMEs. Using the VME index concept, abundance model outputs and previously published presence‐only model predictions were weighted using vulnerability scores, to explore how modelled outputs could provide spatial estimates of likely VME distribution. Spatial predictions of abundance improved on previous modelling to provide an almost complete suite of abundance models for VME indicator taxa in the western portion of the SPRFMO Convention area. Nevertheless, to improve utility of modelled outputs, we recommend more high‐quality seafloor imagery data be gathered within the SPRFMO Convention area to (1) validate abundance models developed here with independent data from the model area, (2) update models, if necessary, (3) link abundance information to ecosystem function and (4) explore validity of the adapted VME index approach used here.

Deep evidential learning for radiotherapy dose prediction

Background:As we navigate towards integrating deep learning methods in the real clinic, a safety concern lies in whether and how the model can express its own uncertainty when making predictions. In this work, we present a novel application of an uncertainty-quantification framework called Deep Evidential Learning in the domain of radiotherapy dose prediction. Method:Using medical images of the Open Knowledge-Based Planning Challenge dataset, we found that this model can be effectively harnessed to yield uncertainty estimates that inherited correlations with prediction errors upon completion of network training. This was achieved only after reformulating the original loss function for a stable implementation. Results:We found that (i)epistemic uncertainty was highly correlated with prediction errors, with various association indices comparable or stronger than those for Monte-Carlo Dropout and Deep Ensemble methods, (ii)the median error varied with uncertainty threshold much more linearly for epistemic uncertainty in Deep Evidential Learning relative to these other two conventional frameworks, indicative of a more uniformly calibrated sensitivity to model errors, (iii)relative to epistemic uncertainty, aleatoric uncertainty demonstrated a more significant shift in its distribution in response to Gaussian noise added to CT intensity, compatible with its interpretation as reflecting data noise. Conclusion:Collectively, our results suggest that Deep Evidential Learning is a promising approach that can endow deep-learning models in radiotherapy dose prediction with statistical robustness. We have also demonstrated how this framework leads to uncertainty heatmaps that correlate strongly with model errors, and how it can be used to equip the predicted Dose-Volume-Histograms with confidence intervals.

Real-time modeling of transient crustal deformation through the quantification of uncertainty deduced from GNSS data

We propose a new method for real-time uncertainty monitoring of earthquake and volcano source models using data from the global navigation satellite system (GNSS) and explore its application concerning observation station placement. The Geospatial Information Authority of Japan operates two main types of GNSS earth observation network system (GEONET) coordinates for crustal deformation monitoring on different time scales: post-processing analysis values and real-time GEONET analysis system for rapid deformation monitoring (REGARD). REGARD uses the Markov Chain Monte Carlo (MCMC) method termed real-time automatic uncertainty estimation of a coseismic single rectangular model using GNSS data (RUNE) for single rectangular fault model estimation to handle uncertainty. Thus far, no GNSS monitoring system can automatically detect transient crustal deformation events, such as volcanic activity and earthquake swarms, on timescales of a day or less. We extended RUNE and developed a core program for a new monitoring system for earthquake and volcanic source models and their uncertainties. Our program achieved automatic and stable MCMC utilization for rectangular fault, dike, Mogi, and spheroid models by increasing the computational speed, improving search efficiency, and adjusting hyperparameters. The program automatically determines the standard deviation of the likelihood function assuming a normal distribution with weights for each observation station. The calculation time was within 15 s for 1 × 106 samples on a standard 1U server. We assessed the reliability of the developed method using synthetic and observed GNSS data from the 2015 Sakurajima volcanic event. The results were consistent with the assumed model and previous studies and indicated an advantage in automatically quantifying uncertainty in a short computation time. Based on MCMC samples, we developed a new visualization algorithm to indicate areas on a map in which the number of observation stations should be expanded. We assessed the reliability using data from the 2023 Noto Peninsula earthquake [Mj 6.5]. The results indicate that the algorithm is helpful in studying the placement of stations. The above model extensions and their application are essential to achieve a rapid quantitative understanding of disaster events near urban areas and for utilizing this information in emergency response activities.Graphical

3Density estimation using spatial capture‐recapture analyses: application to vaccination of prairie dogs against sylvatic plague

AbstractPrairie dogs are notoriously difficult to enumerate, with previously methods including visual counts, mark‐resight, burrow counts, and catch per unit effort. Unlike those methods, spatial capture‐recapture (SCR) analyses allow for formal estimation of density along with associated estimates of uncertainty, detection probability, and the size of the average area over which an individual was detected during the study period (referred to as an activity center). Using SCR analyses, we compared density estimates as part of a field trial evaluating the effectiveness of an oral sylvatic plague vaccine in black‐tailed prairie dogs (Cynomys ludovicianus), Gunnison's prairie dogs (C. gunnisoni), white‐tailed prairie dogs (C. leucurus), and Utah prairie dogs (C. parvidens) at 11 study areas in the western United States. The study was designed as a matched pairs analysis that included 27 individual paired plots (54 plots), each consisting of a plot treated with vaccine baits and a plot treated with placebo baits. Overall, we captured >3,000 individuals each year on these plots, and recapture rates ranged from 5–87%. For black‐tailed prairie dogs, density estimates ranged from 2.7 individuals/ha (95% CI = 2.2–3.3/ha) to 77.3/ha (63.2–94.4/ha), and for Gunnison's prairie dogs, estimates ranged from 11.7/ha (10.6–12.8/ha) to 15.4/ha (14.4–16.7/ha). White‐tailed prairie dogs were at their lowest density (3.3/ha, 95% CI = 2.9–3.8/ha) during the first year of the study and their highest density (14.5/ha; 13.5–15.6/ha) during the last year of the study. Utah prairie dog density estimates ranged from a low of 4.0/ha (95% CI = 3.55–4.6/ha) to a high of 20.8/ha (16.8–25.8/ha). Best‐fitting models of prairie dog density indicated increasing patterns of density over time on most study plots, negative effects of plague, and positive effects of vaccination. Finally, we found low correlations between catch per unit effort estimates from previous published literature at these sites and our densities estimates. Spatial capture‐recapture estimates allowed us to consistently compare treatment effects across space and time, although some exceptions are noted where we observed significant movement between plots within a pair (3 pairs) and when trapping effort between plots or years was not consistent.

Harnessing Representative Spatial-Temporal Information for Video Question Answering

Video question answering, aiming to answer a natural language question related to the given video, has become prevalent in the past few years. Although remarkable improvements have been obtained, it is still exposed to the challenge of insufficient comprehension of video content. To this end, we propose a spatial-temporal representative visual exploitation network for video question answering, which enhances the understanding of the video by merely summarizing representative visual information. In order to explore representative object information, we advance adaptive attention based on uncertainty estimation. At the same time, to capture representative frame-level and clip-level visual information, we structure a much more compact set of representations iteratively in an expectation-maximization manner to deprecate noisy information. Both the quantitative and qualitative results on NExT-QA, TGIF-QA, MSRVTT-QA, and MSVD-QA datasets demonstrate the superiority of our model over several state-of-the-art approaches.

USING RETROSPECTIVE DATA TO ASSESS THE UNCERTAINTY OF INDICATORS OF BIOLOGICAL SAFETY OF FOOD AND FEED RAW MATERIALS

The results of the analysis of the advantages and disadvantages of the use of retrospective data for the implementation of the evaluation of uncertainty in the measurement of indicators of biological safety of food and feed raw materials are given. The model measurement procedure used quantitative determination of genetically modified organisms by means of polymerase chain reaction in real time 250 measurement results of certified reference material with a content of 0.1% genetically modified soybean line obtained under conditions of intra-laboratory reproducibility and 10 results of relevant inter-laboratory comparisons with the participation accredited provider. The suitability of retrospective data was assessed by the Shapiro-Wilk criterion and statistical controllability analysis using standardized score charts and modified Shewhart charts, uncertainty by 4 methods described in general and branch international guidelines and regulations. The results of using a combination of modified Shewhart maps demonstrate the acceptability of this approach for establishing the suitability of retrospective measurement results as input data for uncertainty assessment. By integrating additional "fitness limits" to a set of standardized control limits, this analysis acquires the ability to simultaneously establish not only stability, through the interpretation of geometric patterns, but also compliance with specific criteria. Using this approach, objective evidence of the suitability of retrospective data for estimating measurement uncertainty was obtained. The distribution of retrospective data obtained under conditions of intralaboratory reproducibility is assumed to be normal. As a result of the assessment of statistical controllability and compliance with the established criteria, objective evidence of the suitability of the data for uncertainty assessment was obtained. Analysis of uncertainty intervals of subgroups with the volume of 25 measurements, using 4 assessment methods, demonstrated significant variability in the results of measuring biological safety indicators. The uncertainty interval could narrow or widen by almost 100% of the value within fifty measurements. Median values obtained using different uncertainty estimation methods were 2.69%, 22.69%, 39.23, and 43.08%. This variability is due to the limited coverage of sources of uncertainty, which is inherent in each of the analyzed methods. The most optimal method is the evaluation based on the routine measurement of the certified reference material. Retrospective results of routine measurement of certified reference material and interlaboratory comparisons are an effective source of input data for assessing the uncertainty of the results of measuring biological safety indicators of agricultural raw materials. There is a need for further research to establish an assessment model that will provide an opportunity to take into account the optimal number of influential sources of uncertainty that are inherent in biological systems and cannot be simultaneously taken into account using existing approaches offered by standard assessment methods.

Sommerfeld enhancement at NLO and the dark matter unitarity bound

We reexamine the consequences of perturbative unitarity on dark matter freeze-out when both Sommerfeld enhancement and bound state formation affect dark matter annihilations. At leading order the annihilation cross section is infrared dominated and the connection between the unitarity bound and the upper bound on the dark matter mass depends only on how the different partial waves are populated. We compute how this picture is modified at next-to-leading order with the goal of assigning a reliable theory uncertainty to the freeze-out predictions. We explicitly compute next-to-leading order corrections in a simple model with Abelian gauge interactions and provide an estimate of the theoretical uncertainty for the thermal masses of heavy electroweak n-plets. Along the way, we clarify the regularization and matching procedure necessary to deal with singular potentials in quantum mechanics with a calculable, relativistic UV completion. Published by the American Physical Society 2024

Temperature effect on seawater fCO2 revisited: theoretical basis, uncertainty analysis and implications for parameterising carbonic acid equilibrium constants

Abstract. The sensitivity of the fugacity of carbon dioxide in seawater (fCO2) to temperature (denoted υ, reported in % °C−1) is critical for the accurate fCO2 measurements needed to build global carbon budgets and for understanding the drivers of air–sea CO2 flux variability across the ocean. However, understanding and computing υ have been restricted to either using empirical functions fitted to experimental data or determining it as an emergent property of a fully resolved marine carbonate system, and these two approaches are not consistent with each other. The lack of a theoretical basis and an uncertainty estimate for υ has hindered resolving this discrepancy. Here, we develop a new approach for calculating the temperature sensitivity of fCO2 based on the equations governing the marine carbonate system and the van 't Hoff equation. This shows that, to first order, ln (fCO2) should be proportional to 1/tK (where tK is temperature in kelvin), rather than to temperature, as has previously been assumed. This new approach is, to first order, consistent with calculations from a fully resolved marine carbonate system, which we have incorporated into the PyCO2SYS software. Agreement with experimental data is less convincing but remains inconclusive due to the scarcity of direct measurements of υ, particularly above 25 °C. However, the new approach is consistent with field data, performing better than any other approach for adjusting fCO2 by up to 10 °C if spatiotemporal variability in its single fitted coefficient is accounted for. The uncertainty in υ arising from only measurement uncertainty in the main experimental dataset where υ has been directly measured is in the order of 0.04 % °C−1, which corresponds to a 0.04 % uncertainty in fCO2 adjusted by +1 °C. However, spatiotemporal variability in υ is several times greater than this, so the true uncertainty due to the temperature adjustment in fCO2 adjusted by +1 °C using the most widely used constant υ value is around 0.24 %. This can be reduced to around 0.06 % using the new approach proposed here, and this could be further reduced by more measurements. The spatiotemporal variability in υ arises mostly from the equilibrium constants for CO2 solubility and carbonic acid dissociation (K1∗ and K2∗), and its magnitude varies significantly depending on which parameterisation is used for K1∗ and K2∗. Seawater fCO2 can be measured accurately enough that additional experiments should be able to detect spatiotemporal variability in υ and distinguish between different parameterisations for K1∗ and K2∗. Because the most widely used constant υ was coincidentally measured from seawater with roughly global average υ, our results are unlikely to significantly affect global air–sea CO2 flux budgets, but they may have more important implications for regional budgets and studies that adjust by larger temperature differences.

Advancing EEG prediction with deep learning and uncertainty estimation

Deep Learning (DL) has the potential to enhance patient outcomes in healthcare by implementing proficient systems for disease detection and diagnosis. However, the complexity and lack of interpretability impede their widespread adoption in critical high-stakes predictions in healthcare. Incorporating uncertainty estimations in DL systems can increase trustworthiness, providing valuable insights into the model’s confidence and improving the explanation of predictions. Additionally, introducing explainability measures, recognized and embraced by healthcare experts, can help address this challenge. In this study, we investigate DL models’ ability to predict sex directly from electroencephalography (EEG) data. While sex prediction have limited direct clinical application, its binary nature makes it a valuable benchmark for optimizing deep learning techniques in EEG data analysis. Furthermore, we explore the use of DL ensembles to improve performance over single models and as an approach to increase interpretability and performance through uncertainty estimation. Lastly, we use a data-driven approach to evaluate the relationship between frequency bands and sex prediction, offering insights into their relative importance. InceptionNetwork, a single DL model, achieved 90.7% accuracy and an AUC of 0.947, and the best-performing ensemble, combining variations of InceptionNetwork and EEGNet, achieved 91.1% accuracy in predicting sex from EEG data using five-fold cross-validation. Uncertainty estimation through deep ensembles led to increased prediction performance, and the models were able to classify sex in all frequency bands, indicating sex-specific features across all bands.

Towards a Better Uncertainty Quantification in Automated Valuation Models

AbstractThis study introduces a novel framework for quantifying prediction uncertainty in automated valuation models (AVMs), crucial tools in modern real estate finance. While non-linear AVMs excel in predictive performance, their limited methods for assessing prediction uncertainty reduces reliability and practical utility. We address this gap by proposing an approach for quantifying the uncertainty associated with predicted house prices and by introducing a model-specific AVM uncertainty estimate (AVMU) for AVM comparisons. Using a dataset of 51,747 historical apartment transactions in Oslo, Norway, we train three AVMs (XGBoost, random forest, support vector machine) to predict sales prices. Thereafter, we develop three base uncertainty estimators (direct loss estimation, bootstrap ensemble, quantile regression) and three meta estimators (average regressor, voting regressor, stacked generalization) for uncertainty quantification. Conformal calibration aligns the outputted uncertainty estimates from the six estimators with standard deviations of corresponding prediction errors. Having strong positive correlations with observed absolute prediction errors, the calibrated uncertainty estimators are shown to effectively capture prediction uncertainty. While the direct loss estimation excels among base estimators, the voting regressor and stacked generalization meta estimators consistently outperform it. Furthermore, by using the AVMU estimate from the stacked generalization meta estimator we can successfully identify the best-performing AVM for three separate apartment portfolios without knowing true sales prices. This alignment of the mean estimated prediction uncertainty with observed deviations underlines the utility of pre-factual AVMU estimates for model comparisons. In conclusion, our framework helps bridge prediction accuracy and uncertainty for AVMs, enhancing their reliability and supporting informed decision making for stakeholders.

No evidence for interstellar fireballs in the CNEOS database

Context. The detection of interstellar meteors, especially meteorite-dropping meteoroids, would be transformative, as this would enable direct sampling of material from other stellar systems on Earth. One candidate is the fireball observed by U.S. government sensors on January 8, 2014. It has been claimed that fragments of this meteoroid have been recovered from the ocean floor near Papua New Guinea and that they support an extrasolar origin. Based on its parameters reported in the Center for Near Earth Object Studies (CNEOS) catalog, the fireball exhibits a hyperbolic excess velocity that indicates an interstellar origin; however, the catalog does not report parameter uncertainties. Aims. To achieve a clear confirmation of the fireball’s interstellar origin, we assessed the underlying error distributions of the catalog data. Our aim was also to confirm whether the fragments of this meteoroid survived passage through the atmosphere and assess all conditions needed to unambiguously determine the fragments’ origin. Methods. We approached the investigation of the entire catalog using statistical analyses and modeling, and we provide a comprehensive analysis of the individual hyperbolic CNEOS cases. Results. We have developed several independent arguments indicating substantial uncertainties in the velocity and radiant position of the CNEOS events. We determined that all the hyperbolic fireballs exhibit significant deviations from the majority of the events in one of their velocity components, and we show that such mismeasurements can produce spurious parameters. According to our estimation of the speed measurement uncertainty for the catalog, we found that it is highly probable that such a catalog containing only Sun-bound meteors would show at least one event that appears highly unlikely to be Sun-bound. We also establish that it is unlikely that any fragments from a fireball traveling at the high inferred velocities could survive passage through the atmosphere. When assuming a much lower velocity, some fragments of this meteoroid could survive; however, they would be of a common Solar System origin and thus highly probable to be indistinguishable from the quantity of other local micrometeorites that have gradually accumulated on the sea floor. Conclusions. We conclude that there is no evidence in the CNEOS data to confirm or reject the interstellar origin of any of the nominally hyperbolic fireballs in the CNEOS catalog. Therefore, the claim of an interstellar origin for the fireball recorded over Papua New Guinea in 2014 remains unsubstantiated. We have also gathered arguments that refute the claim that the collected spherules from the sea floor originated in the body of this fireball.

Graph neural processes for molecules: an evaluation on docking scores and strategies to improve generalization

Neural processes (NPs) are models for meta-learning which output uncertainty estimates. So far, most studies of NPs have focused on low-dimensional datasets of highly-correlated tasks. While these homogeneous datasets are useful for benchmarking, they may not be representative of realistic transfer learning. In particular, applications in scientific research may prove especially challenging due to the potential novelty of meta-testing tasks. Molecular property prediction is one such research area that is characterized by sparse datasets of many functions on a shared molecular space. In this paper, we study the application of graph NPs to molecular property prediction with DOCKSTRING, a diverse dataset of docking scores. Graph NPs show competitive performance in few-shot learning tasks relative to supervised learning baselines common in chemoinformatics, as well as alternative techniques for transfer learning and meta-learning. In order to increase meta-generalization to divergent test functions, we propose fine-tuning strategies that adapt the parameters of NPs. We find that adaptation can substantially increase NPs' regression performance while maintaining good calibration of uncertainty estimates. Finally, we present a Bayesian optimization experiment which showcases the potential advantages of NPs over Gaussian processes in iterative screening. Overall, our results suggest that NPs on molecular graphs hold great potential for molecular property prediction in the low-data setting.Scientific contributionNeural processes are a family of meta-learning algorithms which deal with data scarcity by transferring information across tasks and making probabilistic predictions. We evaluate their performance on regression and optimization molecular tasks using docking scores, finding them to outperform classical single-task and transfer-learning models. We examine the issue of generalization to divergent test tasks, which is a general concern of meta-learning algorithms in science, and propose strategies to alleviate it.

TCP-UQ: Two-color Pyrometry Uncertainty Quantification using High Speed Color Cameras

Abstract In this study, an uncertainty quantification method is developed for two color pyrometry (TCP-UQ),
using high-speed color cameras. A NIST traceable calibrated light source is used and the relative exposure
is determined using spectrometer measurements and color camera response functions. The inherent digitization and noise uncertainty is estimated from a set of calibration images and characterized with probability
distribution functions. These uncertainties are then propagated during TCP to two quantities of interest:
the temperature and soot volume fraction measurements. TCP-UQ is applied to a polymethyl methacrylate
(PMMA) slab burner experiment to determine the temperature and the soot volume fraction fields. Temperature measurements ranged from 2300K to 2800K, soot volume fractions are recorded between 0.5 and 10
ppm and are in good agreement with prior measurements reported in the literature. The newly developed
TCP-UQ allows for uncertainty estimations in the temperature and the soot volume fraction measurements
to be associated with variations in the pixel intensities. For the slab burner experiment, the average uncertainty in the temperature is between 120-140 K and the average uncertainty in the soot volume fraction is
around 0.1 ppm. Generally, higher pixel intensities corresponded to lower uncertainty in the temperature
measurement.

Estimating Uncertainty in Actinide Intake Due to Unknown Time of Exposure

Bioassay of radiation workers handling actinides in powder form is carried out annually under routine monitoring program. In case of routine monitoring, as the day of intake is unknown, it is assumed that intake has occurred at the midpoint of monitoring interval. However, an uncertainty is introduced in intake estimation due to this assumption. In the present work, uncertainties were evaluated due to unknown time of intake of Type M and S compounds of Pu and U for annual and biennial monitoring interval. Lung and excreta monitoring methods were considered for uncertainty estimation along with the spread of ±1 month or ±2 months in measurements, at the end of monitoring interval. Annual lung monitoring is found to be the most robust and accurate method for assessment of intake of Type S compounds of Pu, Am and U when time of intake is unknown and Pu is detected using 241Am as a tracer. For Type M compounds, uncertainty was found to be higher than 3 for biennial monitoring interval. Uncertainty is found to be less than 3 for Type S compounds of Pu and U and also for Type M compounds of Pu for both, annual and biennial monitoring intervals using urine analysis. It is higher than 3 for Type M compounds of U for both the monitoring intervals. It is greater than 3 for both Type M and S compounds of Pu and U for both the monitoring intervals using fecal analysis.

Robust quantum dots charge autotuning using neural network uncertainty

Abstract This study presents a machine-learning-based procedure to automate the charge tuning of semiconductor spin qubits with minimal human intervention, addressing one of the significant challenges in scaling up quantum dot technologies. This method exploits artificial neural networks to identify noisy transition lines in stability diagrams, guiding a robust exploration strategy leveraging neural networks' uncertainty estimations. Tested across three distinct offline experimental datasets representing different single quantum dot technologies, the approach achieves over 99% tuning success rate in optimal cases, where more than 10% of the success is directly attributable to uncertainty exploitation. The challenging constraints of small training sets containing high diagram-to-diagram variability allowed us to evaluate the capabilities and limits of the proposed procedure.