Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “variations”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 379 records · Page 21

SU-F-T-609: Impact of Dosimetric Variation for Prescription Dose Using Analytical Anisotropic Algorithm (AAA) in Lung SBRT

Purpose: Actual irradiated prescription dose to patients cannot be verified. Thus, independent dose verification and second treatment planning system are used as the secondary check. AAA dose calculation engine has contributed to lung SBRT. We conducted a multi-institutional study to assess variation of prescription dose for lung SBRT when using AAA in reference to using Acuros XB and Clarkson algorithm. Methods: Six institutes in Japan participated in this study. All SBRT treatments were planed using AAA in Eclipse and Adaptive Convolve (AC) in Pinnacle3. All of the institutes used a same independent dose verification software program (Simple MU Analysis: SMU, Triangle Product, Ishikawa, Japan), which implemented a Clarkson-based dose calculation algorithm using CT image dataset. A retrospective analysis for lung SBRT plans (73 patients) was performed to compute the confidence limit (CL, Average±2SD) in dose between the AAA and the SMU. In one of the institutes, a additional analysis was conducted to evaluate the variations between the AAA and the Acuros XB (AXB). Results: The CL for SMU shows larger systematic and random errors of 8.7±9.9 % for AAA than the errors of 5.7±4.2 % for AC. The variations of AAA correlated with the mean CT values in the voxels of PTV (a correlation coefficient : −0.7) . The comparison of AXB vs. AAA shows smaller systematic and random errors of −0.7±1.7%. The correlation between dose variations for AXB and the mean CT values in PTV was weak (0.4). However, there were several plans with more than 2% deviation of AAPM TG114 (Maximum: −3.3 %). Conclusion: In comparison for AC, prescription dose calculated by AAA may be more variable in lung SBRT patient. Even AXB comparison shows unexpected variation. Care should be taken for the use of AAA in lung SBRT. This research is partially supported by Japan Agency for Medical Research and Development (AMED)

60 APPLIED LIFE SCIENCES↗

Projective Integral Updates for High-Dimensional Variational Inference

Variational inference is an approximation framework for Bayesian inference that seeks to improve quantified uncertainty in predictions by optimizing a simplified distribution over parameters to stand in for the full posterior. Capturing model variations that remain consistent with training data enables more robust predictions by reducing parameter sensitivity. This work introduces a fixed-point optimization for variational inference that is applicable when every feasible log density can be expressed as a linear combination of functions from a given basis. In such cases, the optimizer becomes a fixed-point of projective integral updates. When the basis spans univariate quadratics in each parameter, the feasible distributions are Gaussian mean-fields and the projective integral updates yield quasi-Newton variational Bayes (QNVB). Other bases and updates are also possible. Since these updates require high-dimensional integration, this work begins by proposing an efficient quasirandom sequence of quadratures for mean-field distributions. Each iterate of the sequence contains two evaluation points that combine to correctly integrate all univariate quadratic functions and, if the mean-field factors are symmetric, all univariate cubics. More importantly, averaging results over short subsequences achieves periodic exactness on a much larger space of multivariate polynomials of quadratic total degree. The corresponding variational updates require four loss evaluations with standard (not second-order) backpropagation to eliminate error terms from over half of all multivariate quadratic basis functions. Furthermore, this integration technique is motivated by first proposing stochastic blocked mean-field quadratures, which may be useful in other contexts. A PyTorch implementation of QNVB allows for better control over model uncertainty during training than competing methods. Experiments demonstrate superior generalizability for multiple learning problems and architectures.

Gaussian mean-field↗

Spatial variation of void ratio and shear band thickness in sand using X-ray computed tomography

A detailed study of spatial variation of void ratio and shear band thickness measurements is presented in this paper. A prismatic sand specimen that initially measured 57{center_dot}4 mm wide x 120{center_dot}5 mm long x 182{center_dot}1 mm high was sheared under plane-strain (biaxial) loading conditions. The specimen was prepared at a relative density of 79% using F-75 Ottawa sand. X-ray computed tomography was used to scan the specimen before shearing and after the onset of the shear band. The specimen failed through a single shear band with an inclination angle of 65{center_dot}6{sup o} measured from the direction of the minor principal stress. Computer algorithms were developed to calibrate CT images and quantify void ratio (e) variation within the specimen. CT data analysis revealed cross-sectional spatial variation in void ratio where density is higher in regions close to the edges of the specimen due to membrane effects and confining pressure with no significant variation in void ratio in the axial direction (top to bottom) before shearing. The shear band was easily identified from the CT images, and analysis of void ratio showed a noticeable jump in void ratio profile across the shear band. A detailed statistical summary of the thickness of the shear band and variation of void ratio across the shear band is presented and discussed.

58 GEOSCIENCES↗

The relative variational model - A topological view of matter and its properties: Space occupancy by the atoms

Formal definitions of convergence, connected-ness and continuity were established to characterize and describe the crystalline solid and its properties as a unified notion in the topological space. In this unified notion, physical and material properties are modeled by means of an intrinsic and invariable form function: the Relative Variational Model. The crystalline solid is assumed an empty space that has been filled with atoms and phonons, i.e., the crystal is built with packages of matter and energy in a regular and orderly repetitive pattern along three orthogonal dimensions of the space. The spatial occupation of the atom in the crystalline structure is determined by its mean vibrational volume, which also defines the lattice parameter or interatomic distance. However, as packages of vibrational energy, phonons can only exist as vibrations of atoms. Any variation of internal energy is in fact the discretized variations of phonon's population. These variations occur in the quantized modes of vibration, and therefore the balance between the frequency and amplitude of vibrations also is a dynamic variable. In this paper, the Relative Variational Model was applied to de-convolutions of frequency spectra of the inelastic neutron scatterings. Some dynamic aspects of atom vibration were presented and evaluated in support to the model's fundamentals. (authors)

36 MATERIALS SCIENCE↗

Geographic variations in lowermost mantle structure from the ray parameters and decay constants of core-diffracted waves: CMBR Structure From P diff and S diff

We propose an array-based approach for constraining seismic velocity structure in the lowermost mantle by measuring the frequency dependence of the ray parameter p and decay constant γ for core-diffracted waves (P diff and SH diff ). The approach uses an iterative multichannel cross-correlation algorithm that solves for relative arrival times and amplitudes of core-diffracted waveforms from multiple peaks in normalized correlograms from pairs of coazimuthal stations. The approach is applied to 60 m b ≥ 5.8 earthquakes for P diff and 36 for SH diff during 2001–2002, with nearly 50,000 unique profile station pairs with epicentral-distance differences of Δd ≥ 10° and azimuth differences of Δφ ≤ 10°, sampling a significant portion of the lowermost mantle. Cap-averaging the resulting ray parameter estimates produces geographic variations that are largely consistent with the distribution of lowermost mantle large low-velocity provinces and seismically fast slab-accumulation regions seen in seismic tomography models, whose locations also greatly influence the geographic distribution of heat flow out of the core. Geographic variations in station-pair diffracted-wave decay constants differ from those of the ray parameters, suggesting that variations in the decay of core-diffracted waves are more linked to lowermost mantle seismic velocity gradients than absolute values of seismic velocity. The ray parameters and decay constants of core-diffracted waves are strongly frequency-dependent, and these frequency variations also vary significantly with geographic location. Combining lateral and vertical seismic-velocity variations with mineral-physics data on elasticity and conductivity of lowermost mantle species can provide constraints on D" composition and CMB heat flux.

58 GEOSCIENCES↗

Identification of a rare homozygous c.790C>T variation in the TFB2M gene in Korean patients with autism spectrum disorder

Highlights: • TFB2M have been known to critical in mitochondrial DNA gene expression. • We identified a variation of TFB2M gene sequence in Korean autism patients. • The variation caused increased mitochondrial DNA gene expression. • Increased mitochondrial biogenesis resulted in ROS production and damaged cells. • Structural changes of variant TFB2M seems to delay unloading of DNA from TFB2M. Mitochondrial dysfunction and subsequent enhanced oxidative stress is implicated in the pathogenesis of autism spectrum disorder (ASD). Mitochondrial transcription factor B2 (TFB2M) is an essential protein in mitochondrial gene expression. No reports have described TFB2M mutations and variations involved in any human diseases. We identified a rare homozygous c.790C>T (His264Tyr) variation in TFB2M gene in two Korean siblings with ASD by whole-exome sequencing. The roles of the TFB2M variation in the pathogenesis of ASD were investigated. Patient fibroblasts revealed increased transcription of mitochondrial genes and mitochondrial function in terms of ATP, membrane potential, oxygen consumption, and reactive oxygen species (ROS). Overexpression of the TFB2M variant in primary-cultured fibroblasts demonstrated significantly increased transcription of mitochondrial genes and mitochondrial function compared with overexpression of wild-type TFB2M. Molecular dynamics simulation of the TFB2M variant protein suggested an increase in the rigidity of the hinge region, which may cause alterations in loading and/or unloading of TFB2M on target DNA. Our results suggest that augmentation of mitochondrial gene expression and subsequent enhancement of mitochondrial function may be associated with the pathogenesis of ASD in Korean patients.

60 APPLIED LIFE SCIENCES↗

High variations of methanogenic microorganisms drive full-scale anaerobic digestion process

Anaerobic digestion is one of the most successful waste management strategies worldwide, wherein microorganisms play an essential role in reducing organic pollutants and producing renewable energy. However, variations of microbial community in full-scale anaerobic digesters, particularly functional groups relevant to biogas production, remain elusive. Here, we examined microbial community in a year-long monthly time series of 3 full-scale anaerobic digesters. We observed substantial diversification in community composition, with only a few abundant OTUs (e.g. Clostridiales, Anaerolineaceae and Methanosaeta) persistently present across different samples. Similarly, there were high variations in relative abundance of methanogenic archaea and methanogenic genes, which were positively correlated (r 2 = 0.530, P < 0.00 1). Variations of methanogens explained 55.7% of biogas producing rates, much higher than the explanatory percentage of environmental parameters (16.4%). Hydrogenotrophic methanogens, especially abundant Methanomicrobiales taxa, were correlated with biogas production performance (r = 0.665, P < 0.001) and nearly all methanogenic genes (0.430 < r < 0.735, P < 0.012). Given that methanogenic archaea or genes are well established for methanogenesis, we conclude that high variations in methanogenic traits (e.g. taxa or genes) are responsible for biogas production variations in full-scale anaerobic digesters.

59 BASIC BIOLOGICAL SCIENCES↗

Prostate Contouring Variation: Can It Be Fixed?

Purpose: To assess whether an education program on CT and MRI prostate anatomy would reduce inter- and intraobserver prostate contouring variation among experienced radiation oncologists. Methods and Materials: Three patient CT and MRI datasets were selected. Five radiation oncologists contoured the prostate for each patient on CT first, then MRI, and again between 2 and 4 weeks later. Three education sessions were then conducted. The same contouring process was then repeated with the same datasets and oncologists. The observer variation was assessed according to changes in the ratio of the encompassing volume to intersecting volume (volume ratio [VR]), across sets of target volumes. Results: For interobserver variation, there was a 15% reduction in mean VR with CT, from 2.74 to 2.33, and a 40% reduction in mean VR with MRI, from 2.38 to 1.41 after education. A similar trend was found for intraobserver variation, with a mean VR reduction for CT and MRI of 9% (from 1.51 to 1.38) and 16% (from 1.37 to 1.15), respectively. Conclusion: A well-structured education program has reduced both inter- and intraobserver prostate contouring variations. The impact was greater on MRI than on CT. With the ongoing incorporation of new technologies into routine practice, education programs for target contouring should be incorporated as part of the continuing medical education of radiation oncologists.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Dosimetry estimation on variations of patient size in prostate volumetric-modulated arc therapy

This study investigated the dosimetric variations of the target and critical organs of patients who had weight loss associated with prostate volumetric-modulated arc therapy (VMAT). Five patients with prostate volumes ranging from 32–86.5 cm{sup 3} were selected from a group of 30 patients. Prostate VMAT plans were carried out on each patient using the 6-MV photon beam with a single 360° arc. Decrease of patient size as a result of weight loss was mimicked by contracting the patient's external contour in the anterior, left, and right directions with depths from 0.5–2 cm. Soft tissue excluded by the contracted external contour was replaced by air and the dose distribution was recalculated using the same beam geometry and dose prescription. Dose-volume histograms and dose-volume points such as D99% and D5% for the planning target volume (PTV), clinical target volume (CTV), rectum, bladder, and femoral heads were calculated with variations of reduced depth. In addition, the minimum, maximum, and mean doses for the target and critical organs were determined. PTV and CTV D99% were found to have increased 2.86 ± 0.30% per cm and 2.75 ± 0.38% per cm of reduced depth ranging from 0.5–2 cm. Moreover, the rectal and bladder D30% increased 2.20 ± 0.20% per cm and 2.31 ± 0.83% per cm, and the femoral head D5% increased 3.30 ± 0.11% per cm of reduced depth. Results from variations of the minimum, maximum, and mean doses of the PTV, CTV, rectum, bladder, and femoral heads showed that there was a >5% increase of dose when the reduced depth reached 2 cm. This study provided dosimetry estimation for radiation oncology staff to justify dose variations of the target and critical organs when patients' weight loss occurred in prostate VMAT. Dose variations >5% were seen when the patients' reduced depth was equal to 2 cm.

61 RADIATION PROTECTION AND DOSIMETRY↗

A Generalized Variational Principle with Applications to Excited State Mean Field Theory

We present a generalization of the variational principle that is compatible with any Hamiltonian eigenstate that can be specified uniquely by a list of properties. This variational principle appears to be compatible with a wide range of electronic structure methods, including mean field theory, density functional theory, multireference theory, and quantum Monte Carlo. Like the standard variational principle, this generalized variational principle amounts to the optimization of a nonlinear function that, in the limit of an arbitrarily flexible wave function, has the desired Hamiltonian eigenstate as its global minimum. Unlike the standard variational principle, it can target excited states and select individual states in cases of degeneracy or near-degeneracy. As an initial demonstration of how this approach can be useful in practice, we employ it to improve the optimization efficiency of excited state mean field theory by an order of magnitude. With this improved optimization, we are able to demonstrate that the accuracy of the corresponding second-order perturbation theory rivals that of singles-and-doubles equation-of-motion coupled cluster in a substantially broader set of molecules than could be explored by our previous optimization methodology.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

The Diurnal Variation of the Aerosol Optical Depth at the ARM SGP Site

Abstract This study examines the diurnal variation of the aerosol optical depth (AOD) at 355 nm observed by Raman lidar (RL) at the Atmospheric Radiation Measurement Program Southern Great Plains (SGP) site under both clear and cloudy‐sky conditions. Here only cloudy‐skies when the lidar signal is not fully attenuated are considered. The daytime AOD and its variation from the RL showed an excellent agreement with the Aerosol Robotic Network, demonstrating that the RL‐retrieved AOD is not affected by solar background contamination. The climatological annual‐mean daytime‐mean AOD is only slightly larger than the nighttime‐mean AOD (by 1%–3%). However, day‐to‐day variations are observed such that the daytime‐ and nighttime‐mean AOD difference for a given day can be large (about 95% of days have differences within 0.2). The seasonal AOD diurnal range (i.e., the difference between the maximum and minimum values) relative to the mean was 10%–15% except in the winter when it was 44%. The seasonal‐mean cloudy‐sky AOD diurnal variation is similar to that for clear‐sky, except that the AODs are larger (the annual‐mean cloudy‐sky AOD is larger than the clear‐sky by 24%). The aerosol lidar ratio diurnal variations are also examined, which are 10%–20% for all seasons with a minimum near 9 a.m. to 15 p.m. for all seasons except winter. Also presented is the annual‐mean AOD from the Cloud‐Aerosol Lidar and Infrared Pathfinder Satellite at SGP site: its daytime AOD is about 0.1 smaller than nighttime AOD because of daytime solar background contamination.

54 ENVIRONMENTAL SCIENCES↗

Hair Proteome Variation at Different Body Locations on Genetically Variant Peptide Detection for Protein-Based Human Identification

Abstract Human hair contains minimal intact nuclear DNA for human identification in forensic and archaeological applications. In contrast, proteins offer a pathway to exploit hair evidence for human identification owing to their persistence, abundance, and derivation from DNA. Individualizing single nucleotide polymorphisms (SNPs) are often conserved as single amino acid polymorphisms in genetically variant peptides (GVPs). Detection of GVP markers in the hair proteome via high-resolution tandem mass spectrometry permits inference of SNPs with known statistical probabilities. To adopt this approach for forensic investigations, hair proteomic variation and its effects on GVP identification must first be characterized. This research aimed to assess variation in single-inch head, arm, and pubic hair, and discover body location-invariant GVP markers to distinguish individuals. Comparison of protein profiles revealed greater body location-specific variation in keratin-associated proteins and intracellular proteins, allowing body location differentiation. However, robust GVP markers derive primarily from keratins that do not exhibit body location-specific differential expression, supporting GVP identification independence from hair proteomic variation at the various body locations. Further, pairwise comparisons of GVP profiles with 8 SNPs demonstrated greatest interindividual variation and high intraindividual consistency, enabling similar differentiative potential of individuals using single hairs irrespective of body location origin.

59 BASIC BIOLOGICAL SCIENCES↗

Climate-driven divergence in plant-microbiome interactions generates range-wide variation in bud break phenology

Soil microbiomes are rapidly becoming known as an important driver of plant phenotypic variation and may mediate plant responses to environmental factors. However, integrating spatial scales relevant to climate change with plant intraspecific genetic variation and soil microbial ecology is difficult, making studies of broad inference rare. Here we hypothesize and show: 1) the degree to which tree genotypes condition their soil microbiomes varies by population across the geographic distribution of a widespread riparian tree, Populus angustifolia; 2) geographic dissimilarity in soil microbiomes among populations is influenced by both abiotic and biotic environmental variation; and 3) soil microbiomes that vary in response to abiotic and biotic factors can change plant foliar phenology. We show soil microbiomes respond to intraspecific variation at the tree genotype and population level, and geographic variation in soil characteristics and climate. Using a fully reciprocal plant population by soil location feedback experiment, we identified a climate-based soil microbiome effect that advanced and delayed bud break phenology by approximately 10 days. These results demonstrate a landscape-level feedback between tree populations and associated soil microbial communities and suggest soil microbes may play important roles in mediating and buffering bud break phenology with climate warming, with whole ecosystem implications.

59 BASIC BIOLOGICAL SCIENCES↗

Spatially resolving variations in giant magnetoresistance, undetectable with four-point probe measurements, using infrared microspectroscopy

Magnetorefractive infrared (IR) microspectroscopy is demonstrated to resolve spatial variations in giant magnetoresistance (GMR) and, by modelling, provide an insight into the origin of the variations. Spatial variations are shown to be masked in conventional four-point probe electrical or IR spectral measurements. IR microspectroscopy was performed at the SMIS beamline at the SOLEIL synchrotron, modified to enable measurements in magnetic fields. A GMR gradient was induced in a CoFe/Cu multilayer sample by annealing in a temperature gradient. Modelling revealed that variations in GMR at 900 Oe could be attributed to local variations in interlayer coupling locally changing the switching field.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Near-surface density profiling of Fe ion irradiated Si (100) using extremely asymmetric x-ray diffraction by variation of the wavelength

In this work, we report on correlations between surface density variations and ion parameters during ion beam-induced surface patterning process. The near-surface density variations of irradiated Si(100) surfaces were investigated after off-normal irradiation with 5 keV Fe ions at different fluences. In order to reduce the x-ray probing depth to a thickness below 5 nm, the extremely asymmetrical x-ray diffraction by variation of wavelength was applied, exploiting x-ray refraction at the air-sample interface. Depth profiling was achieved by measuring x-ray rocking curves as function of varying wavelengths providing incidence angles down to 0°. The density variation was extracted from the deviations from kinematical Bragg angle at grazing incidence angles due to refraction of the x-ray beam at the air-sample interface. The simulations based on the dynamical theory of x-ray diffraction revealed that while a net near-surface density decreases with increasing ion fluence which is accompanied by surface patterning, there is a certain threshold of ion fluence to surface density modulation. Our finding suggests that the surface density variation can be relevant with the mechanism of pattern formation.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Degenerate variational integrators for magnetic field line flow and guiding center trajectories

Symplectic integrators offer many benefits for numerically approximating solutions to Hamiltonian differential equations, including bounded energy error and the preservation of invariant sets. Two important Hamiltonian systems encountered in plasma physics—the flow of magnetic field lines and the guiding center motion of magnetized charged particles—resist symplectic integration by conventional means because the dynamics are most naturally formulated in non-canonical coordinates. New algorithms were recently developed using the variational integration formalism; however, those integrators were found to admit parasitic mode instabilities due to their multistep character. This work eliminates the multistep character, and therefore the parasitic mode instabilities via an adaptation of the variational integration formalism that we deem “degenerate variational integration.” Both the magnetic field line and guiding center Lagrangians are degenerate in the sense that the resultant Euler-Lagrange equations are systems of first-order ordinary differential equations. We show that retaining the same degree of degeneracy when constructing discrete Lagrangians yields one-step variational integrators preserving a non-canonical symplectic structure. Lastly, numerical examples demonstrate the benefits of the new algorithms, including superior stability relative to the existing variational integrators for these systems and superior qualitative behavior relative to non-conservative algorithms.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

On the statistical theory of self-gravitating collisionless dark matter flow: Scale and redshift variation of velocity and density distributions

The statistics of velocity and density fields are crucial for cosmic structure formation and evolution. Here, this paper extends our previous work on the two-point second-order statistics for the velocity field [Phys. Fluids 35, 077105 (2023)] to one-point probability distributions for both density and velocity fields. The scale and redshift variation of density and velocity distributions are studied by a halo-based non-projection approach. First, all particles are divided into halo and out-of-halo particles so that the redshift variation can be studied via generalized kurtosis of distributions for halo and out-of-halo particles, respectively. Second, without projecting particle fields onto a structured grid, the scale variation is analyzed by identifying all particle pairs on different scales $r$. We demonstrate that: (i) Delaunay tessellation can be used to reconstruct the density field. The density correlation, spectrum, and dispersion functions were obtained, modeled, and compared with the N-body simulation; (ii) the velocity distributions are symmetric on both small and large scales and are non-symmetric with a negative skewness on intermediate scales due to the inverse energy cascade on small scales with a constant rate $\varepsilon_u$; (iii) On small scales, the even order moments of pairwise velocity $\Delta u_L$ follow a two-thirds law $\propto{(-\varepsilon_ur)}^{2/3}$, while the odd order moments follow a linear scaling $\langle(\Delta u_L)^{2n+1}\rangle=(2n+1)\langle(\Delta u_L)^{2n}\rangle\langle\Delta u_L\rangle\propto{r}$; (iv) The scale variation of the velocity distributions was studied for longitudinal velocities $u_L$ or $u_L^{'}$, pairwise velocity (velocity difference) $\Delta u_L$=$u_L^{'}$-$u_L$ and velocity sum $\Sigma u_L$=$u^{'}_L$+$u_L$. Fully developed velocity fields are never Gaussian on any scale, despite that they can initially be Gaussian; (v) On small scales, $u_L$ and $\Sigma u_L$ can be modeled by a $X$ distribution to maximize the entropy of the system. The distribution of $\Delta u_L$ can be different; (vi) On large scales, $\Delta u_L$ and $\Sigma u_L$ can be modeled by a logistic or a $X$ distribution, while $u_L$ has a different distribution; (vii) the redshift variation of the velocity distributions follows the evolution of the $X$ distribution involving a shape parameter $\alpha(z)$ decreasing with time.

79 ASTRONOMY AND ASTROPHYSICS↗

The Sensitivity of Variational Bayesian Neural Network Performance to Hyperparameters

In scientific applications, predictive modeling is often of limited use without accurate uncertainty quantification (UQ) to indicate when a model may be extrapolating or when more data needs to be collected. Bayesian Neural Networks (BNNs) produce predictive uncertainty by propagating uncertainty in neural network (NN) weights and offer the promise of obtaining not only an accurate predictive model but also accurate UQ. However, in practice, obtaining accurate UQ with BNNs is difficult due in part to the approximations used for model training (such as those made in variational inference) and in part to the need to choose a suitable set of hyperparameters; these hyperparameters outnumber those needed for traditional NNs and often have opaque effects on the results. We aim to shed light on the effects of hyperparameter choices for variational BNNs by performing a global sensitivity analysis of variational BNN performance under varying hyperparameter settings. Our results indicate that many of the hyperparameters interact with each other to affect both predictive accuracy and UQ. For improved usage of variational BNNs in real-world applications, we suggest that thorough hyperparameter tuning, including tuning of prior hyperparameters and loss function parameters, is essential for accurate UQ in variational BNNs.

97 MATHEMATICS AND COMPUTING↗