Spurious Artifacts Research Articles

A Shifted Boundary Method for the compressible Euler equations

The Shifted Boundary Method (SBM) is applied to compressible Euler flows, with and without shock discontinuities. The SBM belongs to the class of unfitted (or immersed, or embedded) finite element methods and avoids integration over cut cells (and the associated implementation/stability issues) by reformulating the original boundary value problem over a surrogate (approximate) computational domain. Accuracy is maintained by modifying the original boundary conditions using Taylor expansions. Hence the name of the method, that shifts the location and values of the boundary conditions. We specifically discuss the advantages the proposed method offers in avoiding spurious numerical artifacts in two scenarios: (a) when curved boundaries are represented by body-fitted polygonal approximations and (b) when the Kutta condition needs to be imposed in immersed simulations of airfoils. An extensive suite of numerical tests is included.

Denoising of dual-VENC PC-MRI with large high/low VENC ratios.

Dual velocity encoding PC-MRI can produce spurious artifacts when using high ratios of velocityencoding values (VENCs), limiting its ability to generate high-quality images across a wide range of encoding velocities. This study aims to propose and compare dual-VENC correction methods for such artifacts. Two denoising approaches based on spatiotemporal regularization are proposed and compared with a state-of-the-art method based on sign correction. Accuracy is assessed using simulated data from an aorta and brain aneurysm, as well as 8 two-dimensional (2D) PC-MRI ascending aorta datasets. Two temporal resolutions (30,60)ms and noise levels (9,12)dB are considered, with noise added to the complex magnetization. The error is evaluated with respect to the noise-free measurement in the synthetic case and to the unwrapped image without additional noise in the volunteer datasets. In all studied cases, the proposed methods are more accurate than the Sign Correction technique. Using simulated 2D+T data from the aorta (60ms, 9dB), the Dual-VENC (DV) error is reduced to: (Sign Correction); and (proposed techniques). The methods are found to be significantly different (p-value ). Importantly, brain aneurysm data revealed that the Sign Correction method is not suitable, as it increases error when the flow is not unidirectional. All three methods improve the accuracy of in vivo data. The newly proposed methods outperform the Sign Correction method in improving dual-VENC PC-MRI images. Among them, the approach based on temporal differences has shown the highest accuracy.

Machine-Learning Metacomputing for Materials Science Data

Abstract Materials science requires the collection and analysis of great quantities of data. These data almost invariably require various post-acquisition computation to remove noise, classify observations, fit parametric models, or perform other operations. Recently developed machine-learning (ML) algorithms have demonstrated great capability for performing many of these operations, and often produce higher quality output than traditional methods. However, it has been widely observed that such algorithms often suffer from issues such as limited generalizability and the tendency to “over fit” to the input data. In order to address such issues, this work introduces a metacomputing framework capable of systematically selecting, tuning, and training the best available machine-learning model in order to process an input dataset. In addition, a unique “cross-training” methodology is used to incorporate underlying physics or multiphysics relationships into the structure of the resultant ML model. This metacomputing approach is demonstrated on four example problems: repairing “gaps” in a multiphysics dataset, improving the output of electron back-scatter detection crystallographic measurements, removing spurious artifacts from X-ray microtomography data, and identifying material constitutive relationships from tensile test data. The performance of the metacomputing framework on these disparate problems is discussed, as are future plans for further deploying metacomputing technologies in the context of materials science and mechanical engineering.

Motion-Aware Video Frame Interpolation

Video frame interpolation methodologies endeavor to create novel frames betwixt extant ones, with the intent of augmenting the video’s frame frequency. However, current methods are prone to image blurring and spurious artifacts in challenging scenarios involving occlusions and discontinuous motion. Moreover, they typically rely on optical flow estimation, which adds complexity to modeling and computational costs. To address these issues, we introduce a Motion-Aware Video Frame Interpolation (MA-VFI) network, which directly estimates intermediate optical flow from consecutive frames by introducing a novel hierarchical pyramid module. It not only extracts global semantic relationships and spatial details from input frames with different receptive fields, enabling the model to capture intricate motion patterns, but also effectively reduces the required computational cost and complexity. Subsequently, a cross-scale motion structure is presented to estimate and refine intermediate flow maps by the extracted features. This approach facilitates the interplay between input frame features and flow maps during the frame interpolation process and markedly heightens the precision of the intervening flow delineations. Finally, a discerningly fashioned loss centered around an intermediate flow is meticulously contrived, serving as a deft rudder to skillfully guide the prognostication of said intermediate flow, thereby substantially refining the precision of the intervening flow mappings. Experiments illustrate that MA-VFI surpasses several representative VFI methods across various datasets, and can enhance efficiency while maintaining commendable efficacy.

Color matching in the wild

We present a method that, given two different views of the same scene taken by two cameras with unknown settings and internal parameters, corrects the colors of one of the images making it look as if it was captured under the other camera settings. Our method is able to deal with any standard non-linear encoded images (gamma-corrected, logarithmic-encoded, or any other) without requiring any previous knowledge of the encoding.To this end, our method makes use of two important observations. First, the camera imaging pipeline from RAW to sRGB can be well approximated by considering just a per-pixel shading and a color transformation matrix, and second, for correcting the images we only need to estimate a single matrix –that will contain information from both of the original images– and an approximation of the shading term (that emulates the non-linearity).Our proposed method is fast and the results have no spurious artifacts. The method outperforms the state-of-the-art when compared with other methods that do not require knowledge of the encoding used. It is also able to compete with –and even surpass in some cases– methods that consider information about image encoding.

Well balanced pick-up assembly for double pulsed magnetic field apparatus

The generation of high magnetic pulsed fields involves several technological challenges. State-of-the-art field peak intensities can be achieved by means of double-coil systems and, as for other configurations, a special pick-up system is required to clean the signal from spurious components and artifacts. However, the design of a well-balanced pick-up for double-coil systems is not a trivial task because of the different mutual inductance of the two field coils. In general the pick-up can be optimized for one coil only, but not for both. Here we present a new pick-up concept specifically aimed at solving this problem, based on an active compensation bridge and a special compensating coil located outside field-generating coils. Besides yielding a good balance of the signal over all pulse duration, a further advantage is an additional increase in the signal-to-noise ratio.

A universal method for solar filament detection from H-alpha observations using semi-supervised deep learning

o he Observatory for Solar and Environmental Research (KSO; Austria) provide regular full-disk observations of the Sun in the core of the chromospheric Halpha absorption line. In this paper, we present a deep learning method that provides reliable extractions of solar filaments from Halpha filtergrams. First, we trained the object detection algorithm YOLOv5 with labeled filament data of ChroTel Halpha filtergrams. We used the trained model to obtain bounding boxes from the full GONG archive. In a second step, we applied a semi-supervised training approach where we used the bounding boxes of filaments to train the algorithm on a pixel-wise classification of solar filaments with u-net. We made use of the increased data set size, which avoids overfitting of spurious artifacts from the generated training masks. Filaments were predicted with an accuracy of 92<!PCT!>. With the resulting filament segmentations, physical parameters such as the area or tilt angle could be easily determined and studied. We demonstrated this in an example where we determined the rush-to-the pole for Solar Cycle 24 from the segmented GONG images. In a last step, we applied the filament detection to Halpha observations from KSO and demonstrated the general applicability of our method to Halpha filtergrams.

Multiphysics Missing Data Synthesis: A Machine Learning Approach for Mitigating Data Gaps and Artifacts

Abstract The presence of gaps and spurious nonphysical artifacts in datasets is a nearly ubiquitous problem in many scientific and engineering domains. In the context of multiphysics numerical models, data gaps may arise from lack of coordination between modeling elements and limitations of the discretization and solver schemes employed. In the case of data derived from physical experiments, the limitations of sensing and data acquisition technologies, as well as myriad sources of experimental noise, may result in the generation of data gaps and artifacts. In the present work, we develop and demonstrate a machine learning (ML) meta-framework for repairing such gaps in multiphysics datasets. A unique “cross-training” methodology is used to ensure that the ML models capture the underlying multiphysics of the input datasets, without requiring training on datasets free of gaps/artifacts. The general utility of this approach is demonstrated by the repair of gaps in a multiphysics dataset taken from hypervelocity impact simulations. Subsequently, we examine the problem of removing scan artifacts from X-ray computed microtomographic (XCMT) datasets. A unique experimental methodology for acquiring XCMT data, wherein articles are scanned multiple times under different conditions, enables the ready identification of artifacts, their removal from the datasets, and the filling of the resulting gaps using the ML framework. This work concludes with observations regarding the unique features of the developed methodology, and a discussion of potential future developments and applications for this technology.

Evaluating spatially enabled machine learning approaches to depth to bedrock mapping, Alberta, Canada.

Maps showing the thickness of sediments above the bedrock (depth to bedrock, or DTB) are important for many geoscience studies and are necessary for many hydrogeological, engineering, mining, and forestry applications. However, it can be difficult to accurately estimate DTB in areas with varied topography, like lowland and mountainous terrain, because traditional methods of predicting bedrock elevation often underestimate or overestimate the elevation in rugged or incised terrain. Here, we describe a machine learning spatial prediction approach that uses information from traditional digital elevation model derived estimates of terrain morphometry and satellite imagery, augmented with spatial feature engineering techniques to predict DTB across Alberta, Canada. First, compiled measurements of DTB from borehole lithologs were used to train a natural language model to predict bedrock depth across all available lithologs, significantly increasing the dataset size. The combined data were then used for DTB modelling employing several algorithms (XGBoost, Random forests, and Cubist) and spatial feature engineering techniques, using a combination of geographic coordinates, proximity measures, neighbouring points, and spatially lagged DTB estimates. Finally, the results were contrasted with DTB predictions based on modelled relationships with the auxiliary variables, as well as conventional spatial interpolations using inverse-distance weighting and ordinary kriging methods. The results show that the use of spatially lagged variables to incorporate information from the spatial structure of the training data significantly improves predictive performance compared to using auxiliary predictors and/or geographic coordinates alone. Furthermore, unlike some of the other tested methods such as using neighbouring point locations directly as features, spatially lagged variables did not generate spurious spatial artifacts in the predicted raster maps. The proposed method is demonstrated to produce reliable results in several distinct physiographic sub-regions with contrasting terrain types, as well as at the provincial scale, indicating its broad suitability for DTB mapping in general.

Deep Learning Model for Quality Assessment of Urinary Bladder Ultrasound Images using Multi-scale and Higher-order Processing.

Autonomous Ultrasound Image Quality Assessment (US-IQA) is a promising tool to aid the interpretation by practicing sonographers and to enable the future robotization of ultrasound procedures. However, autonomous US-IQA has several challenges. Ultrasound images contain many spurious artifacts, such as noise due to handheld probe positioning, errors in the selection of probe parameters and patient respiration during the procedure. Further, these images are highly variable in appearance with respect to the individual patient's physiology. We propose to use a deep Convolutional Neural Network (CNN), USQNet, which utilizes a Multi-scale and Local-to-Global Second-order Pooling (MS-L2GSoP) classifier to conduct the sonographer-like assessment of image quality. This classifier first extracts features at multiple scales to encode the inter-patient anatomical variations, similar to a sonographer's understanding of anatomy. Then, it uses second-order pooling in the intermediate layers (local) and at the end of the network (global) to exploit the second-order statistical dependency of multi-scale structural and multi-region textural features. The L2GSoP will capture the higher-order relationships between different spatial locations and provide the seed for correlating local patches, much like a sonographer prioritizes regions across the image. We experimentally validated the USQNet for a new dataset of the human urinary bladder ultrasound images. The validation involved first with the subjective assessment by experienced radiologists' annotation, and then with state-of-the-art CNN networks for US-IQA and its ablated counterparts. The results demonstrate that USQNet achieves a remarkable accuracy of 92.4% and outperforms the SOTA models by 3 - 14% while requiring comparable computation time.

Unequal-mass boson-star binaries: initial data and merger dynamics

We present a generalisation of the curative initial data construction derived for equal-mass compact binaries in Helfer et al (2019 Phys. Rev. D 99 044046; 2022 Class. Quantum Grav. 39 074001) to arbitrary mass ratios. We demonstrate how these improved initial data avoid substantial spurious artifacts in the collision dynamics of unequal-mass boson-star binaries in the same way as has previously been achieved with the simpler method restricted to the equal-mass case. We employ the improved initial data to explore in detail the impact of phase offsets in the coalescence of equal- and unequal-mass boson star binaries.

Impact of gradient scheme and non-linear shimming on out-of-voxel echo artifacts in edited MRS.

Out-of-voxel (OOV) signals are common spurious echo artifacts in MRS. These signals often manifest in the spectrum as very strong "ripples," which interfere with spectral quantification by overlapping with targeted metabolite resonances. Dephasing optimization through coherence order pathway selection (DOTCOPS) gradient schemes are algorithmically optimized to suppress all potential alternative coherence transfer pathways (CTPs), and should suppress unwanted OOV echoes. In addition, second-order shimming uses non-linear gradient fields to maximize field homogeneity inside the voxel, which unfortunately increases the diversity of local gradient fields outside of the voxel. Given that strong local spatial B0 gradients can refocus unintended CTPs, it is possible that OOVs are less prevalent when only linear first-order shimming is applied. Here we compare the size of unwanted OOV signals in Hadamard-edited (HERMES) data acquired with either a local gradient scheme (which we refer to here as "Shared") or DOTCOPS, and with first- or second-order shimming. We collected data from 15 healthy volunteers in two brain regions (voxel size 30 × 26 × 26 mm3 ) from which it is challenging to acquire MRS data: medial prefrontal cortex and left temporal cortex. Characteristic OOV echoes were seen in both GABA- and GSH-edited spectra for both brain regions, gradient schemes, and shimming approaches. A linear mixed-effect model revealed a statistically significant difference in the average residual based on the gradient scheme in both GABA- (p< 0.001) and GSH-edited (p< 0.001) spectra: that is, the DOTCOPS gradient scheme resulted in smaller OOV artifacts compared with the Shared scheme. There were no significant differences in OOV artifacts associated with shimming method. Thus, these results suggest that the DOTCOPS gradient scheme for J-difference-edited PRESS acquisitions yields spectra with smaller OOV echo artifacts than the Shared gradient scheme implemented in a widely disseminated editing sequence.

Improving cosmological covariance matrices with machine learning

Cosmological covariance matrices are fundamental for parameter inference, since they are responsible for propagating uncertainties from the data down to the model parameters. However, when data vectors are large, in order to estimate accurate and precise covariance matrices we need huge numbers of observations, or rather costly simulations - neither of which may be viable. In this work we propose a machine learning approach to alleviate this problem in the context of the covariance matrices used in the study of large-scale structure. With only a small amount of data (matrices built with samples of 50-200 halo power spectra) we are able to provide significantly improved covariance matrices, which are almost indistinguishable from the ones built from much larger samples (thousands of spectra). In order to perform this task we trained convolutional neural networks to denoise the covariance matrices, using in the training process a data set made up entirely of spectra extracted from simple, inexpensive halo simulations (mocks). We then show that the method not only removes the noise in the covariance matrices of the cheap simulation, but it is also able to successfully denoise the covariance matrices of halo power spectra from N-body simulations. We compare the denoised matrices with the noisy sample covariance matrices using several metrics, and in all of them the denoised matrices score significantly better, without any signs of spurious artifacts. With the help of the Wishart distribution we show that the end product of the denoiser can be compared with an effective sample augmentation in the input matrices. Finally, we show that, by using the denoised covariance matrices, the cosmological parameters can be recovered with nearly the same accuracy as when using covariance matrices built with a sample of 30,000 spectra in the case of the cheap simulations, and with 15,000 spectra in the case of the N-body simulations. Of particular interest is the bias in the Hubble parameter H 0, which was significantly reduced after applying the denoiser.

Compressed sensing with a jackknife and a bootstrap

Compressed sensing proposes to reconstruct more degrees of freedom in a signal than the number of values actually measured (based on a potentially unjustified regularizer or prior distribution). Compressed sensing therefore risks introducing errors -- inserting spurious artifacts or masking the abnormalities that medical imaging seeks to discover. Estimating errors using the standard statistical tools of a jackknife and a bootstrap yields "error bars" in the form of full images that are remarkably qualitatively representative of the actual errors (at least when evaluated and validated on data sets for which the ground truth and hence the actual error is available). These images show the structure of possible errors -- without recourse to measuring the entire ground truth directly -- and build confidence in regions of the images where the estimated errors are small. Further visualizations and summary statistics can aid in the interpretation of such error estimates. Visualizations include suitable colorizations of the reconstruction, as well as the obvious "correction" of the reconstruction by subtracting off the error estimates. The canonical summary statistic would be the root-mean-square of the error estimates. Unfortunately, colorizations appear likely to be too distracting for actual clinical practice in medical imaging, and the root-mean-square gets swamped by background noise in the error estimates. Fortunately, straightforward displays of the error estimates and of the "corrected" reconstruction are illuminating, and the root-mean-square improves greatly after mild blurring of the error estimates; the blurring is barely perceptible to the human eye yet smooths away background noise that would otherwise overwhelm the root-mean-square.

Differential richness inference for 16S rRNA marker gene surveys

BackgroundIndividual and environmental health outcomes are frequently linked to changes in the diversity of associated microbial communities. Thus, deriving health indicators based on microbiome diversity measures is essential. While microbiome data generated using high-throughput 16S rRNA marker gene surveys are appealing for this purpose, 16S surveys also generate a plethora of spurious microbial taxa.ResultsWhen this artificial inflation in the observed number of taxa is ignored, we find that changes in the abundance of detected taxa confound current methods for inferring differences in richness. Experimental evidence, theory-guided exploratory data analyses, and existing literature support the conclusion that most sub-genus discoveries are spurious artifacts of clustering 16S sequencing reads. We proceed to model a 16S survey’s systematic patterns of sub-genus taxa generation as a function of genus abundance to derive a robust control for false taxa accumulation. These controls unlock classical regression approaches for highly flexible differential richness inference at various levels of the surveyed microbial assemblage: from sample groups to specific taxa collections. The proposed methodology for differential richness inference is available through an R package, Prokounter.ConclusionsFalse species discoveries bias richness estimation and confound differential richness inference. In the case of 16S microbiome surveys, supporting evidence indicate that most sub-genus taxa are spurious. Based on this finding, a flexible method is proposed and is shown to overcome the confounding problem noted with current approaches for differential richness inference.Package availability: https://github.com/mskb01/prokounter

NoiER: An Approach for Training More Reliable Fine-Tuned Downstream Task Models

The recent development in pretrained language models trained in a self-supervised fashion, such as BERT, is driving rapid progress in the field of NLP. However, their brilliant performance is based on leveraging syntactic artifacts of the training data rather than fully understanding the intrinsic meaning of language. The excessive exploitation of spurious artifacts causes a problematic issue: The distribution collapse problem, which is the phenomenon that the model fine-tuned on downstream tasks is unable to distinguish out-of-distribution (OOD) sentences while producing a high confidence score. In this paper, we argue that distribution collapse is a prevalent issue in pretrained language models and propose noise entropy regularisation (NoiER) as an efficient learning paradigm that solves the problem without auxiliary models and additional~data. The proposed approach improved traditional OOD detection evaluation metrics by 55% on average compared to the original fine-tuned models.

Opposition Observations of 162173 Ryugu: Normal Albedo Map Highlights Variations in Regolith Characteristics

Abstract On 2019 January 8, the Telescopic Optical Navigation Camera (ONC-T) on board the Hayabusa2 spacecraft observed the Cb-type asteroid 162173 Ryugu under near-opposition illumination and viewing conditions from approximately 20 km in distance. Although opposition observations have never been used for mapping purposes of a planetary body, we found three advantages for mapping under these conditions: (1) images are free of topographic shadows, (2) the reflectance is nearly independent of the orientation of the surface, and (3) spurious color artifacts that may appear near shadowed terrain are avoided. We present normal albedo maps, one for each of the seven filters (0.40–0.95 μm), using an empirical photometric correction. Global coverage of Ryugu is 99.4%. The 0.55 μm band average normal albedo is 4.06% ± 0.10%. Various spectral variations are derived from these maps. Spectral features of regions and boulders are quantified by examining the normal albedo-derived spectral slope and UV index (spectral slope from visible to ultraviolet wavelength) value. In terms of space weathering, three spectral characteristics are observed over the majority of Ryugu: (1) reddening, (2) increases in reflectance at ultraviolet wavelengths compared to visible, and (3) darkening. By contrast, the bright boulders (“type 3”) show a different trend, with wide variations in the 0.95 μm albedo and UV index. Finally, principal component analysis (PCA) comparisons with other asteroids strongly suggest that the main components of Ryugu belong to the B-Cb-type populations. The PCA feature of the fresh material on Ryugu is close to the Eulalia family.

An open-source framework for the implementation of large-scale integral operators with flexible, modern high-performance computing solutions: Enabling 3D Marchenko imaging by least-squares inversion

Numerical integral operators of the convolution type form the basis of most wave-equation-based methods for the processing and imaging of seismic data. Because several of these methods require the solution of an inverse problem, multiple forward and adjoint passes of the modeling operator are generally required to converge to a satisfactory solution. We have highlighted the memory requirements and computational challenges that arise when implementing such operators on 3D seismic data sets and their use for solving large systems of integral equations. A Python framework is presented that leverages libraries for distributed storage and computing, and it provides a high-level symbolic representation of linear operators. A driving goal for our work is not only to offer a widely deployable, ready-to-use high-performance computing framework, but also to demonstrate that it enables addressing research questions that are otherwise difficult to tackle. To this end, the first example of 3D full-wavefield target-oriented imaging, which comprises two subsequent steps of seismic redatuming, is presented. The redatumed fields are estimated by means of gradient-based inversion using the full data set as well as spatially decimated versions of the data set as a way to investigate the robustness of both inverse problems to spatial aliasing in the input data set. Our numerical example shows that when one spatial direction is finely sampled, satisfactory redatuming and imaging can also be accomplished when the sampling in the other direction is coarser than a quarter of the dominant wavelength. Although aliasing introduces noise in the redatumed fields, they are less sensitive to well-known spurious artifacts compared to cheaper, adjoint-based redatuming techniques. These observations are shown to hold for a relatively simple geologic structure, and although further testing is needed for more complex scenarios, we expect them to be generally valid while possibly breaking down for extreme cases (e.g., highly heterogeneous/scattering media).

Explainable Artificial Intelligence for Bias Detection in COVID CT-Scan Classifiers.

Problem: An application of Explainable Artificial Intelligence Methods for COVID CT-Scan classifiers is presented. Motivation: It is possible that classifiers are using spurious artifacts in dataset images to achieve high performances, and such explainable techniques can help identify this issue. Aim: For this purpose, several approaches were used in tandem, in order to create a complete overview of the classificatios. Methodology: The techniques used included GradCAM, LIME, RISE, Squaregrid, and direct Gradient approaches (Vanilla, Smooth, Integrated). Main results: Among the deep neural networks architectures evaluated for this image classification task, VGG16 was shown to be most affected by biases towards spurious artifacts, while DenseNet was notably more robust against them. Further impacts: Results further show that small differences in validation accuracies can cause drastic changes in explanation heatmaps for DenseNet architectures, indicating that small changes in validation accuracy may have large impacts on the biases learned by the networks. Notably, it is important to notice that the strong performance metrics achieved by all these networks (Accuracy, F1 score, AUC all in the 80 to 90% range) could give users the erroneous impression that there is no bias. However, the analysis of the explanation heatmaps highlights the bias.

Oscillation mitigation of hyperbolicity-preserving intrusive uncertainty quantification methods for systems of conservation laws

In this article we study intrusive uncertainty quantification schemes for systems of conservation laws with uncertainty. While intrusive methods inherit certain advantages such as adaptivity and an improved accuracy, they suffer from two key issues. First, intrusive methods tend to show oscillations, especially at shock structures and second, standard intrusive methods can lose hyperbolicity. The aim of this work is to tackle these challenges with the help of two different strategies. First, we combine filters with the multi-element approach for the hyperbolicity-preserving stochastic Galerkin (hSG) scheme. While the limiter used in the hSG scheme ensures hyperbolicity, the filter as well as the multi-element ansatz mitigate oscillations. Second, we derive a multi-element approach for the intrusive polynomial moment (IPM) method. Even though the IPM method is inherently hyperbolic, it suffers from oscillations while requiring the solution of an optimization problem in every spatial cell and every time step. The proposed multi-element IPM method leads to a decoupling of the optimization problem in every multi-element. Thus, we are able to significantly decrease computational costs while improving parallelizability. Both proposed strategies are extended to adaptivity, allowing to adapt the number of basis functions in each multi-element to the smoothness of the solution. We finally evaluate and compare both approaches on various numerical examples such as a NACA airfoil and a nozzle test case for the two-dimensional Euler equations. In our numerical experiments, we observe the mitigation of spurious artifacts. Furthermore, using the multi-element ansatz for IPM significantly reduces computational costs.