Search NASA⌕ Search

SEARCH · Search NASA

Results for “Variational inference”

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 541 records · Page 30

Use of Design of Experiments in Determining Neural Network Architectures for Loss of Control Detection

We describe empirical methods for selecting a neural network architecture to implement belief state inference on generic commercial transport aircraft. We highlight a case study on the planning, execution, and analysis of a set of experiments to determine the configurations of a conditional variational autoencoder (CVAE). Our main contribution is the application of a structured method that can be used for machine learning in many aerospace applications. This method optimizes the structure and training parameters of a neural network for belief state inference, using Design of Experiments (DOE) statistical methodologies. The motivation for this specific DOE analysis was to identify the appropriate hyperparameters for measuring the CVAE reconstruction probability and latent space, such that the measurements can be used to infer qualitative state changes for the aircraft. We demonstrate that this process yields information about a trained neural network’s utility for this specific application, along with a quantifiable range of certainty. We execute 84 experiments using loss-of-control flight maneuver data from the NASA T 2 aircraft, demonstrating that this empirical process allows us to construct cheap and simple models with specific attributes amenable to belief state inference in aerospace applications.

Loss of Control↗

The response of C1O radical concentrations to variations in NO2 radical concentrations in the lower stratosphere

The respose of ClO concentrations to changes in NO2 concentrations has been inferred from simultaneous observations of (ClO), (NO), (NO2) and (O3) in the midlatitude lower stratosphere. This analysis demonstrates that (ClO) is inversely correlated with (NO2), consistent with formation and photolysis of (ClONO2). A factor of ten range in the concentration if NO2 was sampled (0.1 to 1 x 10(exp 9) mol/cu cm), with a comparable range in the ratio of (ClO) to total available inorganic chlorine (1% less than or equal to (ClO)/(Cl(sub y)) less than or equal to 5%. This analysis leads to an estimate of (ClONO2)/(Cl(sub y)) = 0.12 (x/2), in the mid-latitude, lower-stratospheric air masses sampled.

Stimfle, R. M.↗

The Response of ClO Radical Concentrations to Variations in NO2 Radical Concentrations in the Lower Stratosphere

The response of ClO concentrations to changes in NO2 concentrations has been inferred from simultaneous observations of [ClO], [NO], [NO2] and [O3] in the mid-latitude lower stratosphere. This analysis demonstrates that [ClO] is inversely correlated with [NO2], consistent with formation and photolysis of [ClONO2]. A factor of ten range in the concentration of NO2 was sampled (0.1 to 1 x 10(exp 9) mol/cc), with a comparable range in the ratio of [ClO] to total available inorganic chlorine (1% <= [ClO]/[Cl(sub y)] <= 5%). This analysis leads to an estimate of [ClONO2]/[Cl(sub y)] = 0.12 (x/divided by 2), in the mid-latitude, lower-stratospheric air masses sampled.

Stimpfle, R. M.↗

How well was total ozone abundance inferred with Mariner 9?

Ozone is a key to understanding atmospheric chemistry on Mars. Over 20 photochemical models of the martian atmosphere have been published, and O3 is often used as a benchmark for these models. O3 abundance has been inferred from instrumentation on several spacecraft, with the most complete coverage provided by Mariner 9. The Mariner 9 UV spectrometer scanned from 2100 to 3500 Angstroms with a spectral resolution of 15 Angstroms and an effective field-of-view of approximately 300 sq km. The only atmospheric absorption in the 2000 to 3000 A wavelength region was previously assumed to come from the Hartley band system of ozone, which has an opacity of order unity at winter polar latitudes. Therefore, the amount of ozone was inferred by fitting this absorption feature with laboratory data of ozone absorption. Mars O3 shows strong seasonal and latitudinal variation, with column abundances ranging from 0.2 micron-atm at equatorial latitudes to 60 micron-atm over northern winter polar latitudes (1 micron-atm is a column abundance of 2.689 x 10 exp 15 molecules cm(exp -2)). However, the O3 abundance is never great enough to significantly affect atmospheric temperatures or surface temperatures and frost amounts. Some of the previously-inferred O3 abundances are shown. A radiative transfer computer model is used to re-examine the Mariner 9 UV spectra. Assuming a constant mixing ratio for O3 and no chemical or radiative interaction between O3 and clouds/dust, it is shown that when typical amounts of dust and cloud are present that significant underestimation of O3 abundance occurs. A factor of 3 times as much O3 is needed to generate the same spectrum the spacecraft would measure for a cloudy, dusty atmosphere as for a clear atmosphere. If the scattering properties of martian clouds and dust were well known, then their appearance would not be a problem, as a model would be capable of retrieving the O3 abundance. However, these properties are not well known, which raises doubts about the effectiveness of the UV reflectance spectroscopy technique for measuring O3 abundance on Mars.

Lindner, Bernhard Lee↗

Laboratory measurements of microwave absorption from gaseous atmospheric constituents under conditions for the outer planets

Quite often the interpretive work on the microwave and millimeter-wave absorption profiles, which are inferred from radio occultation measurements or radio astronomical observations of the outer planets, employs theoretically-derived absorption coefficients to account for contributions to the observed opacity from gaseous constituents. Variations of the actual absorption coefficients from those which are theoretically derived, especially under the environmental conditions characteristic of the outer planets, can result in significant errors in the inferred abundances of the absorbing constituents. The recognition of the need to make laboratory measurements of the absorptivity of gases such as NH3, CH4, and H2O in a predominantly H2 atmosphere, under temperature and pressure conditions simulating the outer planets' atmospheres, and at wavelengths corresponding to both radio occultation and radio astronomical observations, has led to the development of a facility capable of making such measurements at Georgia Tech. The laboratory measurement system, the measurement techniques, and the proposed experimental regimen for Winter 1985 are described.

Steffes, Paul G.↗

Use of Design of Experiments in Determining Neural Network Architectures for Loss of Control Detection

We describe empirical methods for selecting a neural network architecture to implement belief state inference on generic commercial transport aircraft. We highlight a case study on the planning, execution, and analysis of a set of experiments to determine the configurations of a conditional variational autoencoder (CVAE). Our main contribution is the application of a structured method that can be used for machine learning in many aerospace applications. This method optimizes the structure and training parameters of a neural network for belief state inference, using Design of Experiments (DOE) statistical methodologies. The motivation for this specific DOE analysis was to identify the appropriate hyperparameters for measuring the CVAE reconstruction probability and latent space, such that the measurements can be used to infer qualitative state changes for the aircraft. We demonstrate that this process yields information about a trained neural network’s utility for this specific application, along with a quantifiable range of certainty. We execute 84 experiments using loss-of-control flight maneuver data from the NASA T-2 aircraft, demonstrating that this empirical process allows us to construct cheap and simple models with specific attributes amenable to belief state inference in aerospace applications.

Loss of Control↗

Whole Earth Telescope observations of V471 Tauri - The nature of the white dwarf variations

Time-series photometric observations of the binary star V471 Tauri were conducted using the Whole Earth Telescope observing network. The purpose was to determine the mechanism responsible for causing the 555 and 277 s periodic luminosity variations exhibited by the white dwarf in this binary. Previous observers have proposed that either g-mode pulsations or rotation of an accreting magnetic white dwarf could cause the variations, but were unable to decide which was the correct model. The present observations have answered this question. Learning the cause of the white dwarf variations has been possible because of the discovery of a periodic signal at 562 s in the Johnson U-band flux of the binary. By identifying this signal as reprocessed radiation and using its phase to infer the phase of the shorter wavelength radiation which produces it, made it possible to compare the phase of the 555 s U-band variations to the phase of the X-ray variations. It was found that U-band maximum coincides with X-ray minimum. From this result it was concluded that the magnetic rotator model accurately describes the variations observed, but that models involving g-mode pulsations do not.

Clemens, J. C.↗

Discovery of the third transient X-ray binary in the galactic globular cluster Terzan 5

We report and study the outburst of a new transient X-ray binary (XRB) in Terzan 5, the third detected in this globular cluster, Swift J174805.3-244637 or Terzan 5 X-3. We find clear spectral hardening in Swift/XRT data during the outburst rise to the hard state, thanks to our early coverage (starting at L{sub X} ∼ 4 × 10{sup 34} erg s{sup –1}) of the outburst. This hardening appears to be due to the decline in relative strength of a soft thermal component from the surface of the neutron star (NS) during the rise. We identify a Type I X-ray burst in Swift/XRT data with a long (16 s) decay time, indicative of hydrogen burning on the surface of the NS. We use Swift/BAT, MAXI/GSC, Chandra/ACIS, and Swift/XRT data to study the spectral changes during the outburst, identifying a clear hard-to-soft state transition. We use a Chandra/ACIS observation during outburst to identify the transient's position. Seven archival Chandra/ACIS observations show evidence for variations in Terzan 5 X-3's nonthermal component but not the thermal component during quiescence. The inferred long-term time-averaged mass accretion rate, from the quiescent thermal luminosity, suggests that if this outburst is typical and only slow cooling processes are active in the NS core, such outbursts should recur every ∼10 yr.

79 ASTRONOMY AND ASTROPHYSICS↗

Monitoring the variability of intrinsic absorption lines in quasar spectra , ,

We have monitored 12 intrinsic narrow absorption lines (NALs) in five quasars and seven mini-broad absorption lines (mini-BALs) in six quasars for a period of 4-12 yr (1-3.5 yr in the quasar rest-frame). We present the observational data and the conclusions that follow immediately from them, as a prelude to a more detailed analysis. We found clear variability in the equivalent widths (EWs) of the mini-BAL systems but no easily discernible changes in their profiles. We did not detect any variability in the NAL systems or in narrow components that are often located at the center of mini-BAL profiles. Variations in mini-BAL EWs are larger at longer time intervals, reminiscent of the trend seen in variable BALs. If we assume that the observed variations result from changes in the ionization state of the mini-BAL gas, we infer lower limits to the gas density ∼10{sup 3}-10{sup 5} cm{sup –3} and upper limits on the distance of the absorbers from the central engine of the order of a few kiloparsecs. Motivated by the observed variability properties, we suggest that mini-BALs can vary because of fluctuations of the ionizing continuum or changes in partial coverage while NALs can vary primarily because of changes in partial coverage.

79 ASTRONOMY AND ASTROPHYSICS↗

A MACHINE-LEARNING METHOD TO INFER FUNDAMENTAL STELLAR PARAMETERS FROM PHOTOMETRIC LIGHT CURVES

A fundamental challenge for wide-field imaging surveys is obtaining follow-up spectroscopic observations: there are >10{sup 9} photometrically cataloged sources, yet modern spectroscopic surveys are limited to ∼few× 10{sup 6} targets. As we approach the Large Synoptic Survey Telescope era, new algorithmic solutions are required to cope with the data deluge. Here we report the development of a machine-learning framework capable of inferring fundamental stellar parameters (T {sub eff}, log g, and [Fe/H]) using photometric-brightness variations and color alone. A training set is constructed from a systematic spectroscopic survey of variables with Hectospec/Multi-Mirror Telescope. In sum, the training set includes ∼9000 spectra, for which stellar parameters are measured using the SEGUE Stellar Parameters Pipeline (SSPP). We employed the random forest algorithm to perform a non-parametric regression that predicts T {sub eff}, log g, and [Fe/H] from photometric time-domain observations. Our final optimized model produces a cross-validated rms error (RMSE) of 165 K, 0.39 dex, and 0.33 dex for T {sub eff}, log g, and [Fe/H], respectively. Examining the subset of sources for which the SSPP measurements are most reliable, the RMSE reduces to 125 K, 0.37 dex, and 0.27 dex, respectively, comparable to what is achievable via low-resolution spectroscopy. For variable stars this represents a ≈12%-20% improvement in RMSE relative to models trained with single-epoch photometric colors. As an application of our method, we estimate stellar parameters for ∼54,000 known variables. We argue that this method may convert photometric time-domain surveys into pseudo-spectrographic engines, enabling the construction of extremely detailed maps of the Milky Way, its structure, and history.

79 ASTRONOMY AND ASTROPHYSICS↗

The Completed SDSS-IV extended Baryon Oscillation Spectroscopic Survey: measurement of the BAO and growth rate of structure of the emission line galaxy sample from the anisotropic power spectrum between redshift 0.6 and 1.1

We analyse the large-scale clustering in Fourier space of emission line galaxies (ELG) from the Data Release 16 of the Sloan Digital Sky Survey IV extended Baryon Oscillation Spectroscopic Survey. The ELG sample contains 173,736 galaxies covering 1,170 square degrees in the redshift range 0.6 < z < 1.1. We perform a BAO measurement from the post-reconstruction power spectrum monopole, and study redshift space distortions (RSD) in the first three even multipoles. Furthermore, photometric variations yield fluctuations of both the angular and radial survey selection functions. Those are directly inferred from data, imposing integral constraints which we model consistently. The full data set has only a weak preference for a BAO feature (1.4σ). At the effective redshift zeff = 0.845 we measure $D_{\rm V}(z_{\rm eff})/r_{\rm drag} = 18.33_{-0.62}^{+0.57}$, with DV the volume-averaged distance and rdrag the comoving sound horizon at the drag epoch. In combination with the RSD measurement, at zeff = 0.85 we find $f\sigma _8(z_{\rm eff}) = 0.289_{-0.096}^{+0.085}$, with f the growth rate of structure and σ8 the normalisation of the linear power spectrum, $D_{\rm H}(z_{\rm eff})/r_{\rm drag} = 20.0_{-2.2}^{+2.4}$ and DM(zeff)/rdrag = 19.17 ± 0.99 with DH and DM the Hubble and comoving angular distances, respectively. These results are in agreement with those obtained in configuration space, thus allowing a consensus measurement of fσ8(zeff) = 0.315 ± 0.095, $D_{\rm H}(z_{\rm eff})/r_{\rm drag} = 19.6_{-2.1}^{+2.2}$ and DM(zeff)/rdrag = 19.5 ± 1.0. This measurement is consistent with a flat ΛCDM model with Planck parameters.

79 ASTRONOMY AND ASTROPHYSICS↗

Optical, near, infrared and ultraviolet monitoring of the Seyfert 1 galaxy Markarian 335

Preliminary results of a multifrequency monitoring campaign for the bright, Seyfert 1 galactic nuclei Mkn335 are presented. Nearly uniform sampling at 3 day intervals is achieved quasi simultaneously at each wavelength band. Wavelength dependent variability is seen at the 20 to 30 percent level. Interpretation of variability in terms of geometrically thin, optically thick accretion disk models is discussed. The inferred blackhole masses and accretion rates are discussed. Possible correlation between continuum and emission line variations is discussed.

Shrader, Chris R.↗

The distribution of particulate material on Mars

The surface materials on Mars were extensively studied using a variety of spacecraft and Earth-based remote sensing observations. These measurements include: (1) diurnal thermal measurements, used to determine average particle size, rock abundance, and the presence of crusts; (2) radar observations, used to estimate the surface slope distributions, wavelength scale roughness, and density; (3) radio emission observations, used to estimate subsurface density; (4) broadband albedo measurements, used to study the time variation of surface brightness and dust deposition and removal; and (5) color observations, used to infer composition, mixing, and the presence of crusts. Remote sensing observations generally require some degree of modeling to interpret, making them more difficult to interpret than direct observations from the surface. They do, however, provide a means for examining the surface properties over the entire planet and a means of sampling varying depths within the regolith. Albedo and color observations only indicate the properties of the upper-most few microns, but are very sensitive to thin, sometimes emphemeral dust coatings. Thermal observations sample the upper skin depth, generally 2 to 10 cm. Rock abundance measurements give an indirect indication of surface mantling, where the absence of rocks suggests mantles of several meters. Finally, radar and radio emission data can penetrate several meters into the surface, providing an estimate of subsurface density and roughness.

Christensen, Philip R.↗

Does UV instrumentation effectively measure ozone abundance?

Measurements of O3 on Mars provide significant information about the chemistry and composition of the atmosphere, including long-term changes. The most extensive and accurate data were inferred from the Mariner 9 UV spectrometer experiment. Mars O3 shows strong seasonal and latitudinal variation, with column abundances ranging from 0.2 microns at equatorial latitudes to 60 microns over the northern winter polar latitudes (1 micron-atm is a column abundance of 2.689 x 10(exp 15) molecules cm(exp-2)). The Mariner 9 UV spectrometer scanned from 2100 to 3500 Angstroms in one of its two spectral channels every 3 seconds with a spectral resolution of 15 Angstroms and an effective field-of-view of approximately 300 km(exp 2). Measurements were made for almost half a Martian year, with winter and spring in the Northern Hemisphere and summer and fall in the Southern Hemisphere. The detectability limit of the spectrometer was approximately 3 microns of ozone. The UV spectrometer on Mariner 9 was incapable of penetrating the dust during dust storms; the single-scattering albedo and phase function of airborne dust and cloud ice are not known to the degree required to extract the small UV signal reflected up from near the surface. The reflectance spectroscopy technique would also have difficulty detecting the total column abundance of O3 in cases where large dust abundances exist together with the polar hood, especially at high latitudes where large solar zenith angles magnify those optical depths; yet these cases would contain the maximum O3, based on theoretical results. It is quite possible that the maximum O3 column abundance observed by Mariner 9 of 60 microns is common. In fact, larger quantities may exist in some of the colder areas with optically thick clouds and dust. As the Viking period often had more atmospheric dust loading than did that of Mariner 9, the reflectance spectroscopy technique may even have been incapable of detecting the entire O3 column abundance during much of the Mars year that Viking observed, particularly at high latitudes.

Lindner, Bernhard Lee↗

Patterns of seismic activity preceding large earthquakes

A mechanical model of seismic faults is employed to investigate the seismic activities that occur prior to major events. The block-and-spring model dynamically generates a statistical distribution of smaller slipping events that precede large events, and the results satisfy the Gutenberg-Richter law. The scaling behavior during a loading cycle suggests small but systematic variations in space and time with maximum activity acceleration near the future epicenter. Activity patterns inferred from data on seismicity in California demonstrate a regional aspect; increased activity in certain areas are found to precede major earthquake events. One example is given regarding the Loma Prieta earthquake of 1989 which is located near a fault section associated with increased activity levels.

Shaw, Bruce E.↗

Constraining particle acceleration in Sgr A? with simultaneous GRAVITY, Spitzer, NuSTAR, and Chandra observations

We report the time-resolved spectral analysis of a bright near-infrared and moderate X-ray flare of Sgr A⋆. We obtained light curves in the M, K, and H bands in the mid- and near-infrared and in the 2 − 8 keV and 2 − 70 keV bands in the X-ray. The observed spectral slope in the near-infrared band is νLν ∝ ν0.5 ± 0.2; the spectral slope observed in the X-ray band is νLν ∝ ν−0.7 ± 0.5. Using a fast numerical implementation of a synchrotron sphere with a constant radius, magnetic field, and electron density (i.e., a one-zone model), we tested various synchrotron and synchrotron self-Compton scenarios. The observed near-infrared brightness and X-ray faintness, together with the observed spectral slopes, pose challenges for all models explored. We rule out a scenario in which the near-infrared emission is synchrotron emission and the X-ray emission is synchrotron self-Compton. Two realizations of the one-zone model can explain the observed flare and its temporal correlation: one-zone model in which the near-infrared and X-ray luminosity are produced by synchrotron self-Compton and a model in which the luminosity stems from a cooled synchrotron spectrum. Both models can describe the mean spectral energy distribution (SED) and temporal evolution similarly well. In order to describe the mean SED, both models require specific values of the maximum Lorentz factor γmax, which differ by roughly two orders of magnitude. The synchrotron self-Compton model suggests that electrons are accelerated to γmax ∼ 500, while cooled synchrotron model requires acceleration up to γmax ∼ 5 × 104. The synchrotron self-Compton scenario requires electron densities of 1010 cm−3 that are much larger than typical ambient densities in the accretion flow. Furthermore, it requires a variation of the particle density that is inconsistent with the average mass-flow rate inferred from polarization measurements and can therefore only be realized in an extraordinary accretion event. In contrast, assuming a source size of 1 RS, the cooled synchrotron scenario can be realized with densities and magnetic fields comparable with the ambient accretion flow. For both models, the temporal evolution is regulated through the maximum acceleration factor γmax, implying that sustained particle acceleration is required to explain at least a part of the temporal evolution of the flare.

R. Abuter↗

Machine Learning 5G Attack Detection in Programmable Logic

Machine learning-assisted network security may significantly contribute to securing 5G components. However, machine learning network security inference speeds generally require tens to hundreds of milliseconds thereby introducing significant latency in 5G operations. The inference latency can be reduced by deploying the machine learning model to programmable logic in a field programmable gate array (FPGA) at the cost of a small loss in accuracy. In order to quantify this loss, as well as to establish baseline performance inference speeds for programmable logic implementations, this work explores an autoencoder and a ß-variational autoencoder deployed on two different FPGA evaluation boards and compares accuracy and performance against an NVIDIA A100 GPU implementation. A publicly available 5G dataset containing 10 types of attacks along with normal traffic is introduced as part of the evaluation.

97 MATHEMATICS AND COMPUTING↗