Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Variational Bayesian”

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 271 records · Page 15

Insights into Permafrost and Seasonal Active-Layer Dynamics from Ambient Seismic Noise Monitoring

Widespread permafrost thaw in response to changing climate conditions has the potential to dramatically impact ecosystems, infrastructure, and the global carbon budget. Ambient seismic noise techniques allow passive subsurface monitoring that could provide new insights into permafrost vulnerability and active–layer processes. Using nearly 2 years of continuous seismic data recorded near Fairbanks, Alaska, we measured relative velocity variations that showed a clear seasonal cycle reflecting active–layer freeze and thaw. Relative to January 2014, velocities increased up to 3% through late spring, decreased to –8% by late August, then gradually returned to the initial values by the following winter. Velocities responded rapidly (over ~2 to 7 days) to discrete hydrologic events and temperature forcing and indicated that spring snowmelt and infiltration events from summer rainfall were particularly influential in propagating thaw across the site. Velocity increases during the fall zero–curtain captured the re–freezing process and incremental ice formation. Looking across multiple frequency bands (3–30 Hz), negative relative velocities began at higher frequencies earlier in the summer then shifted lower when active–layer thaw deepened, suggesting a potential relationship between frequency and thaw depth; however, this response was dependent on interstation distance. Bayesian tomography returned 2–D time–lapse images identifying zones of greatest velocity reduction concentrated in the western side of the array, providing insight into the spatial variability of thaw progression, soil moisture, and drainage. Furthermore, this study demonstrates the potential of passive seismic monitoring as a new tool for studying site–scale active–layer and permafrost thaw processes at high temporal and spatial resolution.

58 GEOSCIENCES↗

Supervised detection of anomalous light curves in massive astronomical catalogs

The development of synoptic sky surveys has led to a massive amount of data for which resources needed for analysis are beyond human capabilities. In order to process this information and to extract all possible knowledge, machine learning techniques become necessary. Here we present a new methodology to automatically discover unknown variable objects in large astronomical catalogs. With the aim of taking full advantage of all information we have about known objects, our method is based on a supervised algorithm. In particular, we train a random forest classifier using known variability classes of objects and obtain votes for each of the objects in the training set. We then model this voting distribution with a Bayesian network and obtain the joint voting distribution among the training objects. Consequently, an unknown object is considered as an outlier insofar it has a low joint probability. By leaving out one of the classes on the training set, we perform a validity test and show that when the random forest classifier attempts to classify unknown light curves (the class left out), it votes with an unusual distribution among the classes. This rare voting is detected by the Bayesian network and expressed as a low joint probability. Our method is suitable for exploring massive data sets given that the training process is performed offline. We tested our algorithm on 20 million light curves from the MACHO catalog and generated a list of anomalous candidates. After analysis, we divided the candidates into two main classes of outliers: artifacts and intrinsic outliers. Artifacts were principally due to air mass variation, seasonal variation, bad calibration, or instrumental errors and were consequently removed from our outlier list and added to the training set. After retraining, we selected about 4000 objects, which we passed to a post-analysis stage by performing a cross-match with all publicly available catalogs. Within these candidates we identified certain known but rare objects such as eclipsing Cepheids, blue variables, cataclysmic variables, and X-ray sources. For some outliers there was no additional information. Among them we identified three unknown variability types and a few individual outliers that will be followed up in order to perform a deeper analysis.

79 ASTRONOMY AND ASTROPHYSICS↗

Statistical Characterization of Temperature and Pressure Vertical Profiles for the Analysis of Laser Heterodyne Radiometry Data

The statistical analysis of historic pressure and temperature profiles from radiosonde launches for use in the fitting of molecular oxygen line shapes is presented. As the O 2 mixing ratio is nearly constant throughout the lower atmosphere, only variations in pressure and temperature profiles will affect the fit of observed O 2 features in Laser Heterodyne Radiometry (LHR) spectra. Radiosonde temperature and pressure data are extracted from the Integrated Global Radiosonde Archive (IGRA) for a given station, date, and launch time. Data may be extracted for a single launch, for the same date over several years, and/or within a window centered on a target date. The temperature and pressure profiles are further characterized by the statistical variation in coefficients of polynomial fits in altitude. The properties of the probability distributions for each coefficient are used to constrain fits of O 2 line shapes through Nelder–Mead optimization. The refined temperature and pressure profiles are then used in the retrieval of vertically resolved mixing ratios for greenhouse gases (GHGs) measured in the same instrument. In continuous collections, each vertical profile determination may be treated as a Bayesian prior to inform subsequent measurements and provide an estimate of uncertainties.

47 OTHER INSTRUMENTATION↗

CLASH: JOINT ANALYSIS OF STRONG-LENSING, WEAK-LENSING SHEAR, AND MAGNIFICATION DATA FOR 20 GALAXY CLUSTERS

We present a comprehensive analysis of strong-lensing, weak-lensing shear and magnification data for a sample of 16 X-ray-regular and 4 high-magnification galaxy clusters at 0.19≲z≲0.69 selected from Cluster Lensing And Supernova survey with Hubble (CLASH). Our analysis combines constraints from 16-band Hubble Space Telescope observations and wide-field multi-color imaging taken primarily with Suprime-Cam on the Subaru Telescope, spanning a wide range of cluster radii (10″–16′). We reconstruct surface mass density profiles of individual clusters from a joint analysis of the full lensing constraints, and determine masses and concentrations for all of the clusters. We find the internal consistency of the ensemble mass calibration to be ≤5% ± 6% in the one-halo regime (200–2000 kpc h{sup −1}) compared to the CLASH weak-lensing-only measurements of Umetsu et al. For the X-ray-selected subsample of 16 clusters, we examine the concentration–mass (c–M) relation and its intrinsic scatter using a Bayesian regression approach. Our model yields a mean concentration of c|{sub z=0.34}=3.95±0.35 at M{sub 200c} ≃ 14 × 10{sup 14} M{sub ⊙} and an intrinsic scatter of σ(lnc{sub 200c})=0.13±0.06, which is in excellent agreement with Λ cold dark matter predictions when the CLASH selection function based on X-ray morphological regularity and the projection effects are taken into account. We also derive an ensemble-averaged surface mass density profile for the X-ray-selected subsample by stacking their individual profiles. The stacked lensing signal is detected at 33σ significance over the entire radial range ≤4000 kpc h{sup −1}, accounting for the effects of intrinsic profile variations and uncorrelated large-scale structure along the line of sight. The stacked mass profile is well described by a family of density profiles predicted for cuspy dark-matter-dominated halos in gravitational equilibrium, namely, the Navarro–Frenk–White (NFW), Einasto, and DARKexp models, whereas the single power-law, cored isothermal and Burkert density profiles are disfavored by the data. We show that cuspy halo models that include the large-scale two-halo term provide improved agreement with the data. For the NFW halo model, we measure a mean concentration of c{sub 200c}=3.79{sub −0.28}{sup +0.30} at M{sub 200c}=14.1{sub −1.0}{sup +1.0}×10{sup 14} M{sub ⊙}, demonstrating consistency between the complementary analysis methods.

79 ASTRONOMY AND ASTROPHYSICS↗

MODEL-FREE MULTI-PROBE LENSING RECONSTRUCTION OF CLUSTER MASS PROFILES

Lens magnification by galaxy clusters induces characteristic spatial variations in the number counts of background sources, amplifying their observed fluxes and expanding the area of sky, the net effect of which, known as magnification bias, depends on the intrinsic faint-end slope of the source luminosity function. The bias is strongly negative for red galaxies, dominated by the geometric area distortion, whereas it is mildly positive for blue galaxies, enhancing the blue counts toward the cluster center. We generalize the Bayesian approach of Umetsu et al. for reconstructing projected cluster mass profiles, by incorporating multiple populations of background sources for magnification-bias measurements and combining them with complementary lens-distortion measurements, effectively breaking the mass-sheet degeneracy and improving the statistical precision of cluster mass measurements. The approach can be further extended to include strong-lensing projected mass estimates, thus allowing for non-parametric absolute mass determinations in both the weak and strong regimes. We apply this method to our recent CLASH lensing measurements of MACS J1206.2-0847, and demonstrate how combining multi-probe lensing constraints can improve the reconstruction of cluster mass profiles. This method will also be useful for a stacked lensing analysis, combining all lensing-related effects in the cluster regime, for a definitive determination of the averaged mass profile.

79 ASTRONOMY AND ASTROPHYSICS↗

Sequential spectral line analysis for accurate density and temperature diagnosis of laboratory opacity measurements

The accuracy of iron opacity calculated in stellar interiors has been questioned since the discovery of the “solar problem” and the discrepancies between the measured and modeled iron opacity reported in 2015. Experimental opacity benchmarks require accurate temperature and density measurements, which were inferred by analyzing tracer magnesium spectra in those experiments. Could the observed discrepancy be explained by insufficient accuracy in the inferred temperature, density, and their uncertainties? Previous analyses may have yielded biased results due to three limitations: (1) simultaneous multi-line fitting, (2) approximations in line-shape models, and (3) exclusion of certain spectral lines due to insufficient background characterization. Notably, the first issue is a common concern for many inversion methods, including Bayesian inferences. We present a refined analysis method that overcomes these limitations, applied to three categories of iron opacity experiments (Anchor 1, 2, and 3). In particular, the sequential fitting method yields unbiased results with more realistic uncertainties by accounting for line inconsistencies in the parameter uncertainties. The average electron temperature and density values are 162 ± 6 eV and (7.0 ± 1.9) × 10 21 cm −3 for six Anchor 1 experiments, 189 ± 7 eV and (3.4 ± 0.3) × 10 22 cm −3 for 21 Anchor 2 experiments, and 201 ± 6 eV and (4.8 ± 1.1) × 10 22 cm −3 for nine Anchor 3 experiments. These results show ∼4% temperature and ∼20% density reproducibility over a decade, which also aligns with the inferred parameter uncertainties. In conclusion, the resulting temperature and density uncertainties lead to a quasi-continuum iron opacity variation of ±4%–7% for wavelengths below 9.5 Å, which is insufficient to explain the significant model-data discrepancies reported in 2015.

Absorption spectroscopy↗

Explainable and Differentiable Reinforcement Learning for Multi-objective Optimization in Particle Accelerators

Operating particle accelerators involves optimizing multiple goals simultaneously, which can be challenging due to trade-offs among objectives. While evolutionary algorithms like the genetic algorithm (GA) have been used for various Multi-Objective Optimization (MOO) tasks, they are not inherently suited for complex control problems. This talk highlights two variations of Reinforcement Learning (RL) for concurrently optimizing heat load and trip rates at the Continuous Electron Beam Accelerator Facility (CEBAF). The problem involves strict constraints on individual states, actions, and overall energy requirements of the beam. First, this talk highlights how differentiability can be harnessed through a Deep Differentiable Reinforcement Learning (DDRL) approach to address MOO issues within particle accelerators. We examine the DDRL method alongside Model Free Reinforcement Learning (MFRL), GA, and Bayesian Optimization (BO). The performance of these methods is assessed by generating a Pareto-front for two objectives. Our findings indicate that DDRL excels in handling high-dimensional problems more effectively than MFRL, BO, and GA. Next, we will show integration of explainable physics-based constraints into RL algorithms to enhance trans- parency and trust in decision-making processes by enabling users to verify that agents adhere to established physical principles. This surrogate function can be modeled using neural networks or sparse dictionary mod- els. By examining the mathematical form of the learned constraint function, we are able to confirm the agent has learned to use the established physics of each environment provided but the surrogate model. In addi- tion, we find that the introduction of a mathematical functional dictionary based surrogate model enables our reinforcement learning algorithms to reliably converge for difficult high-dimensional accelerator controls environments.

Rajput, Kishansingh [Thomas Jefferson National Acc↗

NO TIMING VARIATIONS OBSERVED IN THIRD TRANSIT OF SNOW-LINE EXOPLANET KEPLER-421b

We observed Kepler-421 during the anticipated third transit of the snow-line exoplanet Kepler-421b in order to constrain the existence and extent of transit timing variations (TTVs). Previously, the Kepler spacecraft only observed two transits of Kepler-421b, leaving the planet’s transit ephemeris unconstrained. Our visible light, time-series observations from the 4.3 m Discovery Channel Telescope were designed to capture pre-transit baseline and the partial transit of Kepler-421b, barring significant TTVs. We use the light curves to assess the probabilities of various transit models using both the posterior odds ratio and the Bayesian Information Criterion, and find that a transit model with no TTVs is favored to 3.6 σ confidence. These observations suggest that Kepler-421b is either alone in its system or is only experiencing minor dynamic interactions with an unseen companion. With the Kepler-421b ephemeris constrained, we calculate future transit times and discuss the opportunity to characterize the atmosphere of this cold, long-period exoplanet via transmission spectroscopy. Our investigation emphasizes the difficulties associated with observing long-period exoplanet transits and the consequences that arise from failing to refine transit ephemerides.

79 ASTRONOMY AND ASTROPHYSICS↗

Broadband Processing in a Noisy Shallow Ocean Environment: A Particle Filtering Approach

Here we report that when a broadband source propagates sound in a shallow ocean the received data can become quite complicated due to temperature-related sound-speed variations and therefore a highly dispersive environment. Noise and uncertainties disrupt this already chaotic environment even further because disturbances propagate through the same inherent acoustic channel. The broadband (signal) estimation/detection problem can be decomposed into a set of narrowband solutions that are processed separately and then combined to achieve more enhancement of signal levels than that available from a single frequency, thereby allowing more information to be extracted leading to a more reliable source detection. A Bayesian solution to the broadband modal function tracking, pressure-field enhancement, and source detection problem is developed that leads to nonparametric estimates of desired posterior distributions enabling the estimation of useful statistics and an improved processor/detector. In conclusion, to investigate the processor capabilities, we synthesize an ensemble of noisy, broadband, shallow-ocean measurements to evaluate its overall performance using an information theoretical metric for the preprocessor and the receiver operating characteristic curve for the detector.

47 OTHER INSTRUMENTATION↗

Development of a framework for sequential Bayesian design of experiments: Application to a pilot-scale solvent-based CO 2 capture process

In this paper, a methodology is developed for sequential design of experiments (SDoE) for process systems and applied to a solvent-based CO 2 capture system. In this approach, the prior knowledge of the system is used to prioritize process data collection at specific operating conditions. These data are then incorporated into a Bayesian inference methodology for updating a stochastic model by refining estimations of its underlying parameters, and the updated model is then used to generate the next set of test runs. Thus, the new knowledge obtained from the data is used to guide subsequent iterations of the experimental runs, ensuring that the overall data collection is maximally informative given that most experimental campaigns, especially at pilot or higher-scale plants, are costly, time-consuming, and resource-limited. The test run objective for this work was to minimize the maximum model prediction uncertainty for key output variables, but the methodology is generic and can be readily applied to other test run objectives. This methodology is applied to an aqueous monoethanolamine (MEA) pilot plant campaign at the National Carbon Capture Center (NCCC) in Wilsonville, Alabama, USA. The SDoE framework was utilized for two iterations, while collecting 18 sets of data representing different process conditions, and this resulted in an overall average reduction in uncertainty of approximately 50% in the prediction of CO 2 capture percentage. Moreover, 11 additional data sets were obtained with variation of absorber packing height for further model validation. This work shows the capability of the SDoE framework to maximize learning given limited resources, allowing for the reduction of model uncertainty, which is of great importance for many applications including reduction of technical risk associated with scale-up and economic analysis.

20 FOSSIL-FUELED POWER PLANTS↗

Accounting for Epistemic Uncertainty in Mission Supportability Assessment: A Necessary Step in Understanding Risk and Logistics Requirements

Future crewed missions to Mars present a maintenance logistics challenge that is unprecedented in human spaceflight. Mission endurance – defined as the time between resupply opportunities – will be significantly longer than previous missions, and therefore logistics planning horizons are longer and the impact of uncertainty is magnified. Maintenance logistics forecasting typically assumes that component failure rates are deterministically known and uses them to represent aleatory uncertainty, or uncertainty that is inherent to the process being examined. However, failure rates cannot be directly measured; rather, they are estimated based on similarity to other components or statistical analysis of observed failures. As a result, epistemic uncertainty – that is, uncertainty in knowledge of the process – exists in failure rate estimates that must be accounted for. Analyses that neglect epistemic uncertainty tend to significantly underestimate risk. Epistemic uncertainty can be reduced via operational experience; for example, the International Space Station (ISS) failure rate estimates are refined using a Bayesian update process. However, design changes may re-introduce epistemic uncertainty. Thus, there is a tradeoff between changing a design to reduce failure rates and operating a fixed design to reduce uncertainty. This paper examines the impact of epistemic uncertainty on maintenance logistics requirements for future Mars missions, using data from the ISS Environmental Control and Life Support System (ECLS) as a baseline for a case study. Sensitivity analyses are performed to investigate the impact of variations in failure rate estimates and epistemic uncertainty on spares mass. The results of these analyses and their implications for future system design and mission planning are discussed.

Owens, Andrew↗

Daily water stable isotopes, transpiration, and matrix potential data for an aspen and engelmann stand in the East River Watershed (version 2)

We provide daily stable isotope (2H & 18O) ratios in soil water and xylem (plant stem) water, as well as the sap flow (transpiration) and the soil's matric potential at a forested site near Gothic, Colorado, in the East River catchment. We measured the stable isotopic composition of the transpiration and the daily transpiration flux sum of three aspen and three engelmann spruce. In both forest stands, we installed a soil profile and measured the soil matric potential at 15, 30, and 60 cm depth as well as the stable isotopes of soil pore water at 5, 10, 30, 60, and 90 cm depths. All isotope measurements were done in situ via vapor probes connected to a cavity ring down spectrometer (Picarro L1240i).We further report the daily meteorological data observed at billy barr near our study site. We also provide for each tree the relative share of root water uptake derived from the isotope measurements via a Bayesian mixing model (MixSIAR).The daily data is provided as a time series in "Iso_MP_Sap_DataDaily_ESSDiveUpload.csv" and the units are provided in "dd.csv"; the location of the instrumented trees and soil profiles are given as latitude and longitude coordinates saved as CSV and KMZ files; and a file-level metadata (flmd.csv) file that lists each file contained in the dataset with associated metadata.The data was gathered to investigate the short-term changes of the water sources (i.e., variation of root water uptake from different soil depths) of the studied subalpine trees.Update 07/16/2025: The relative and absolute plant water uptake depths were grouped to ensure that the MixSIAR model was applied with endmembers that differed in their d2H value by at least 3 permill and at least 1 permill for d18O. Whenever the difference between observed d2H values for two or more probes at neighboring depths was less than 3 permill, we used the average value for the source water endmember. For days at which probes that were not next to each other measurements did not differ at least 3 permill, the average of all probes between these two depths was used as the water source endmember representing the depth range between these two probes.

54 ENVIRONMENTAL SCIENCES↗

Thermal Behavior of Unusual Local-Scale Features on Vesta

On Vesta, the thermal behavior of areas of unusual albedo seen at the local scale can be related to physical properties that can provide information about the origin of those materials. Dawn's Visible and Infrared Mapping Spectrometer (VIR) hyperspectral cubes are used to retrieve surface temperatures and emissivities, with high accuracy as long as temperatures are greater than ~180 K. Data acquired in the Survey phase (23 July through 29 August 2011) show several unusual surface features: 1) high-albedo (bright) and low-albedo (dark) material deposits, 2) spectrally distinct ejecta and pitted materials, 3) regions suggesting finer-grained materials. Some of the unusual dark and bright features were re-observed by VIR in the subsequent High-Altitude Mapping Orbit (HAMO) and Low-Altitude Mapping Orbit (LAMO) phases at increased pixel resolution. In particular, bright and dark surface materials on Vesta, and pitted materials, are currently being investigated by the Dawn team. In this work we present temperature maps and emissivities of several local-scale features that were observed by Dawn under different illumination conditions and different local solar times. To calculate surface temperatures, we applied a Bayesian approach to nonlinear inversion based on the Kirchhoff law and the Planck function, and whose results were compared with those provided by the application of alternative methods. Data from the IR channel of VIR show that bright regions generally correspond to regions with lower thermal emission, i.e. lower temperature, while dark regions correspond to areas with higher thermal emission, i.e. higher temperature. This behavior confirms that many of the dark appearances in the VIS mainly reflect albedo variations, and not, for example, shadowing. During maximum daily insolation, dark features in the equatorial region may rise to temperatures greater than 270 K, while brightest features stop at roughly 258 K for similar local solar times. However, pitted materials, showing relatively low reflectance, have significantly lower temperatures, as a result of differences in composition and/or structure (e.g, average grain size of the surface regolith, porosity, etc.). To complement this work, we provide preliminary values of thermal inertia for some bright and dark features.

Tosi, Federico↗

A Multi-Route Model of Nicotine-Cotinine Pharmacokinetics, Pharmacodynamics and Brain Nicotinic Acetylcholine Receptor Binding in Humans

The pharmacokinetics of nicotine, the pharmacologically active alkaloid in tobacco responsible for addiction, are well characterized in humans. We developed a physiologically based pharmacokinetic/pharmacodynamic model of nicotine pharmacokinetics, brain dosimetry and brain nicotinic acetylcholine receptor (nAChRs) occupancy. A Bayesian framework was applied to optimize model parameters against multiple human data sets. The resulting model was consistent with both calibration and test data sets, but in general underestimated variability. A pharmacodynamic model relating nicotine levels to increases in heart rate as a proxy for the pharmacological effects of nicotine accurately described the nicotine related changes in heart rate and the development and decay of tolerance to nicotine. The PBPK model was utilized to quantitatively capture the combined impact of variation in physiological and metabolic parameters, nicotine availability and smoking compensation on the change in number of cigarettes smoked and toxicant exposure in a population of 10,000 people presented with a reduced toxicant (50%), reduced nicotine (50%) cigarette Across the population, toxicant exposure is reduced in some but not all smokers. Reductions are not in proportion to reductions in toxicant yields, largely due to partial compensation in response to reduced nicotine yields. This framework can be used as a key element of a dosimetry-driven risk assessment strategy for cigarette smoke constituents.

Teeguarden, Justin G↗

Bayesian inference analysis of jet quenching using inclusive jet and hadron suppression measurements

The JETSCAPE Collaboration reports a new determination of the jet transport parameter $\hat{q}$ in the quark-gluon plasma (QGP) using Bayesian inference, incorporating all available inclusive hadron and jet yield suppression data measured in heavy-ion collisions at the BNL Relativistic Heavy Ion Collider (RHIC) and the CERN Large Hadron Collider (LHC). This multi-observable analysis extends the previously published JETSCAPE Bayesian inference determination of $\hat{q}$, which was based solely on a selection of inclusive hadron suppression data. jetscape is a modular framework incorporating detailed dynamical models of QGP formation and evolution, and jet propagation and interaction in the QGP. Virtuality-dependent partonic energy loss in the QGP is modeled as a thermalized weakly coupled plasma, with parameters determined from Bayesian calibration using soft-sector observables. This Bayesian calibration of $\hat{q}$ utilizes active learning, a machine-learning approach, for efficient exploitation of computing resources. The experimental data included in this analysis span a broad range in collision energy and centrality, and in transverse momentum. In order to explore the systematic dependence of the extracted parameter posterior distributions, several different calibrations are reported, based on combined jet and hadron data; on jet or hadron data separately; and on restricted kinematic or centrality ranges of the jet and hadron data. Tension is observed in comparison of these variations, providing new insights into the physics of jet transport in the QGP and its theoretical formulation.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Bayesian, Multifidelity Operator Learning for Complex Engineering Systems–A Position Paper

Abstract Deep learning has significantly improved the state-of-the-art in computer vision and natural language processing, and holds great potential to design effective tools for predicting and simulating complex engineering systems. In particular, scientific machine learning seeks to apply the power of deep learning to scientific and engineering tasks, with operator learning (OL) emerging as a particularly effective tool. OL can approximate nonlinear operators arising in complex engineering systems, making it useful for simulating, designing, and controlling those systems. In this position paper, we provide a comprehensive overview of OL, including its potential applications to complex engineering domains. We cover three variations of OL approaches: deterministic OL for modeling nonautonomous systems, OL with uncertainty quantification (UQ) capabilities, and multifidelity OL. For each variation, we discuss drawbacks and potential applications to engineering, in addition to providing a detailed explanation. We also highlight how multifidelity OL approaches with UQ capabilities can be used to design, optimize, and control engineering systems. Finally, we outline some potential challenges for OL within the engineering domain.

Computer Science↗

Deeply uncertain: comparing methods of uncertainty quantification in deep learning algorithms

We present a comparison of methods for uncertainty quantification (UQ) in deep learning algorithms in the context of a simple physical system. Three of the most common uncertainty quantification methods - Bayesian Neural Networks (BNN), Concrete Dropout (CD), and Deep Ensembles (DE) - are compared to the standard analytic error propagation. We discuss this comparison in terms endemic to both machine learning ("epistemic" and "aleatoric") and the physical sciences ("statistical" and "systematic"). The comparisons are presented in terms of simulated experimental measurements of a single pendulum - a prototypical physical system for studying measurement and analysis techniques. Our results highlight some pitfalls that may occur when using these UQ methods. For example, when the variation of noise in the training set is small, all methods predicted the same relative uncertainty independently of the inputs. This issue is particularly hard to avoid in BNN. On the other hand, when the test set contains samples far from the training distribution, we found that no methods sufficiently increased the uncertainties associated to their predictions. This problem was particularly clear for CD. In light of these results, we make some recommendations for usage and interpretation of UQ methods.

59 BASIC BIOLOGICAL SCIENCES↗

A Preferential Growth Channel for Supermassive Black Holes in Elliptical Galaxies at z ≲ 2

The assembly of stellar and supermassive black hole (SMBH) mass in elliptical galaxies since z ~ 1 can help to diagnose the origins of locally observed correlations between SMBH mass and stellar mass. We therefore construct three samples of elliptical galaxies, one at z ~ 0 and two at 0.7 ≲ z ≲ 2.5, and quantify their relative positions in the M BH -M * plane. Using a Bayesian analysis framework, we find evidence for translational offsets in both stellar mass and SMBH mass between the local sample and both higher-redshift samples. The offsets in stellar mass are small, and consistent with measurement bias, but the offsets in SMBH mass are much larger, reaching a factor of 7 between z ~ 1 and z ~ 0. The magnitude of the SMBH offset may also depend on redshift, reaching a factor of ~20 at z ~ 2. The result is robust against variation in the high- and low-redshift samples and changes in the analysis approach. The magnitude and redshift evolution of the offset are challenging to explain in terms of selection and measurement biases. We conclude that either there is a physical mechanism that preferentially grows SMBHs in elliptical galaxies at z ≲ 2, or that selection and measurement biases are both underestimated, and depend on redshift.

79 ASTRONOMY AND ASTROPHYSICS↗