SEARCH · Search NASA
Results for “Conditional Distribution”
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.
SIDDA: SInkhorn Dynamic Domain Adaptation for image classification with equivariant neural networks
Modern neural networks (NNs) often do not generalize well in the presence of a ‘covariate shift’; that is, in situations where the training and test data distributions differ, but the conditional distribution of classification labels given the data remains unchanged. In such cases, NN generalization can be reduced to a problem of learning more robust, domain-invariant features. Domain adaptation (DA) methods include a broad range of techniques aimed at achieving this; however, these methods have struggled with the need for extensive hyperparameter tuning, which then incurs significant computational costs. In this work, we introduce SInkhorn Dynamic Domain Adaptation (SIDDA), an out-of-the-box DA training algorithm built upon the Sinkhorn divergence, that can achieve effective domain alignment with minimal hyperparameter tuning and computational overhead. We demonstrate the efficacy of our method on multiple simulated and real datasets of varying complexity, including simple shapes, handwritten digits, real astronomical observations, and remote sensing data. These datasets exhibit covariate shifts due to noise, blurring, differences between telescopes, and variations in imaging wavelengths. SIDDA is compatible with a variety of NN architectures, and it works particularly well in improving classification accuracy and model calibration when paired with symmetry-aware equivariant NNs (ENNs). We find that SIDDA consistently enhances the generalization capabilities of NNs, achieving up to a ${\approx}40\%$ improvement in classification accuracy on unlabeled target data, while also providing a more modest performance gain of $\lesssim 1\%$ on labeled source data. We also study the efficacy of DA on ENNs with respect to the varying group orders of the dihedral group DN, and find that the model performance improves as the degree of equivariance increases. Finally, if SIDDA achieves proper domain alignment, it also enhances model calibration on both source and target data, with the most significant gains in the unlabeled target domain—achieving over an order of magnitude improvement in the expected calibration error and Brier score. SIDDA’s versatility across various NN models and datasets, combined with its automated approach to domain alignment, has the potential to significantly advance multi-dataset studies by enabling the development of highly generalizable models.
Estimating Conditional Climate Distributions via Bayesian Transport Maps
Explore the source record for details and available documents.
SIDDA: SInkhorn Dynamic Domain Adaptation
Modern neural networks (NNs) often do not generalize well in the presence of a "covariate shift"; that is, in situations where the training and test data distributions differ, but the conditional distribution of classification labels remains unchanged. In such cases, NN generalization can be reduced to a problem of learning more domain-invariant features. Domain adaptation (DA) methods include a range of techniques aimed at achieving this; however, these methods have struggled with the need for extensive hyperparameter tuning, which then incurs significant computational costs. In this work, we introduce SIDDA, an out-of-the-box DA training algorithm built upon the Sinkhorn divergence, that can achieve effective domain alignment with minimal hyperparameter tuning and computational overhead. We demonstrate the efficacy of our method on multiple simulated and real datasets of varying complexity, including simple shapes, handwritten digits, and real astronomical observations. SIDDA is compatible with a variety of NN architectures, and it works particularly well in improving classification accuracy and model calibration when paired with equivariant neural networks (ENNs). We find that SIDDA enhances the generalization capabilities of NNs, achieving up to a ≈40% improvement in classification accuracy on unlabeled target data. We also study the efficacy of DA on ENNs with respect to the varying group orders of the dihedral group DN, and find that the model performance improves as the degree of equivariance increases. Finally, we find that SIDDA enhances model calibration on both source and target data--achieving over an order of magnitude improvement in the ECE and Brier score. SIDDA's versatility, combined with its automated approach to domain alignment, has the potential to advance multi-dataset studies by enabling the development of highly generalizable models.
Realizability conditions for distributed rc networks
Explore the source record for details and available documents.
The specification of distributed boundary conditions in numerical simulation of the ionosphere
The approach described makes it possible to use boundary conditions which do not have to be associated with any particular point of the solution interval. The approach was applied in a program which solves simultaneously the coupled differential equations describing the concentrations of the O(+) ions, their temperature, the electron temperature and the concentrations of O2(+), NO(+) and N2(+) ions. Several versions of the proposed scheme are described for various kinds of boundary conditions. Attention is given to a given columnar ionic content, a given incoming or outgoing flux, or a given height of the peak concentration.
Estimating Conditional Climate Distributions via Bayesian Transport Maps
Explore the source record for details and available documents.
Electron density and electron temperature measurement in a bi-Maxwellian electron distribution using a derivative method of Langmuir probes
In plasma diagnostics with a single Langmuir probe, the electron temperature T{sub e} is usually obtained from the slope of the logarithm of the electron current or from the electron energy probability functions of current (I)-voltage (V) curve. Recently, Chen [F. F. Chen, Phys. Plasmas 8, 3029 (2001)] suggested a derivative analysis method to obtain T{sub e} by the ratio between the probe current and the derivative of the probe current at a plasma potential where the ion current becomes zero. Based on this method, electron temperatures and electron densities were measured and compared with those from the electron energy distribution function (EEDF) measurement in Maxwellian and bi-Maxwellian electron distribution conditions. In a bi-Maxwellian electron distribution, we found the electron temperature T{sub e} obtained from the method is always lower than the effective temperatures T{sub eff} derived from EEDFs. The theoretical analysis for this is presented.
Bounds on galaxy stochasticity from halo occupation distribution modeling
The joint probability distribution of matter overdensity and galaxy counts in cells is a powerful probe of cosmology, and the extent to which variance in galaxy counts at fixed matter density deviates from Poisson shot noise is not fully understood. The lack of informed bounds on this stochasticity is currently the limiting factor in constraining cosmology with the galaxy–matter probability distribution function (PDF). We investigate stochasticity in the conditional distribution of galaxy counts along lines of sight with fixed matter density, and we present a halo occupation distribution (HOD)-based approach for obtaining plausible ranges for stochasticity parameters. To probe the high-dimensional space of possible galaxy–matter connections, we derive a set of HODs that conserve the galaxies’ linear bias and number density to produce RED M A G I C-like galaxy catalogs within the A BACUS S UMMIT suite of N -body simulations. We study the impact of individual HOD parameters and cosmology on stochasticity and perform a Monte Carlo search in HOD parameter space subject to the constraints on bias and density. In mock catalogs generated by the selected HODs, shot noise in galaxy counts spans both sub-Poisson and super-Poisson values, ranging from 80% to 133% of Poisson variance for cells with mean matter density. Nearly all of the derived HODs show a positive relationship between local matter density and stochasticity. For galaxy catalogs with higher stochasticity, modeling galaxy bias to second order is required for an accurate description of the conditional PDF of galaxy counts at fixed matter density. The presence of galaxy assembly bias also substantially extends the range of stochasticity in the super-Poisson direction. This HOD-based approach leverages degrees of freedom in the galaxy–halo connection to obtain informed bounds on nuisance model parameters and can be adapted to study other parametrizations of shot noise in galaxy counts, in particular to motivate prior ranges on stochasticity for cosmological analyses.
Geodesic least squares regression on information manifolds
We present a novel regression method targeted at situations with significant uncertainty on both the dependent and independent variables or with non-Gaussian distribution models. Unlike the classic regression model, the conditional distribution of the response variable suggested by the data need not be the same as the modeled distribution. Instead they are matched by minimizing the Rao geodesic distance between them. This yields a more flexible regression method that is less constrained by the assumptions imposed through the regression model. As an example, we demonstrate the improved resistance of our method against some flawed model assumptions and we apply this to scaling laws in magnetic confinement fusion.
Cluster infall for mass calibration in the stage-IV era
The outskirts of galaxy clusters present a promising avenue for constraining cluster masses in a way that is robust to the impact of baryonic physics. We assess the accuracy to which the cluster infall regions can be used for cluster mass calibration. Building on previous work, we parametrize the velocity distribution 𝑃(𝑣r,𝑣tan|𝑟,𝑀) of dark matter halos on scales 𝑟 ≥ 5ℎ −1 Mpc as the product of the marginalized distribution 𝑃(𝑣 r |𝑟,𝑀) and the conditional distribution 𝑃(𝑣 tan |𝑣 r ,𝑟,𝑀), calibrating the radial and mass dependence of these distributions in numerical simulations. We then project our model along the line of sight to obtain accurate predictions for the distributions of line-of-sight velocities at a given projected radius and cluster mass 𝑃(𝑣 LOS |𝑅,𝑀), which we can observe with spectroscopic survey data. Furthermore, with our model, we forecast that spectra from the Dark Energy Spectroscopic Instrument can constrain cluster masses with subpercent-level precision, comparable to that of stage-IV weak lensing surveys.
Elliptically-Contoured Tensor-variate Distributions with Application to Image Learning
Statistical analysis of tensor-valued data has largely used the tensor-variate normal (TVN) distribution that may be inadequate for data arising from distributions with heavier or lighter tails. We study a general family of elliptically contoured (EC) TV distributions and derive its characterizations, moments, marginal, and conditional distributions. We describe procedures for maximum likelihood estimation from data that are (1) uncorrelated draws from an EC distribution, (2) from a scale mixture of the TVN distribution, and (3) from an underlying but unknown EC distribution, for which we extend Tyler’s robust estimator. A detailed simulation study highlights the benefits of choosing an EC distribution over the TVN for heavier-tailed data. We develop TV classification rules using discriminant analysis and EC errors and show that they better predict cats and dogs from images in the Animal Faces-HQ dataset than the TVN-based rules. A novel tensor-on-tensor regression and TV analysis of variance (TANOVA) framework under EC errors is also demonstrated to better characterize gender, age, and ethnic origin than the usual TVN-based TANOVA in the celebrated labeled faces of the wild dataset.
A Pseudoreversible Normalizing Flow for Stochastic Dynamical Systems with Various Initial Distributions
Here, we present a pseudoreversible normalizing flow method for efficiently generating samples of the state of a stochastic differential equation (SDE) with various initial distributions. The primary objective is to construct an accurate and efficient sampler that can be used as a surrogate model for computationally expensive numerical integration of SDEs, such as those employed in particle simulation. After training, the normalizing flow model can directly generate samples of the SDE’s final state without simulating trajectories. The existing normalizing flow model for SDEs depends on the initial distribution, meaning the model needs to be retrained when the initial distribution changes. The main novelty of our normalizing flow model is that it can learn the conditional distribution of the state, i.e., the distribution of the final state conditional on any initial state, such that the model only needs to be trained once and the trained model can be used to handle various initial distributions. This feature can provide a significant computational saving in studies of how the final state varies with the initial distribution. Additionally, we propose to use a pseudoreversible network architecture to define the normalizing flow model, which has sufficient expressive power and training efficiency for a variety of SDEs in science and engineering, e.g., in particle physics. We provide a rigorous convergence analysis of the pseudoreversible normalizing flow model to the target probability density function in the Kullback–Leibler divergence metric. Numerical experiments are provided to demonstrate the effectiveness of the proposed normalizing flow model.
Reconciling Chord Length Distributions and Area Distributions for Fields of Fractal Cumulus Clouds
While the total cover of broken cloud fields can in principle be obtained from one-dimensional measurements, the cloud size distribution normally differs between two-dimensional (area) and one-dimensional retrieval (chord length) methods. In this study, we use output from high-resolution Large Eddy Simulations to generate a transfer function between the two. We retrieve chord lengths and areas for many clouds, and plot the one as a function of the other, and vice versa. We find that the cloud area distribution conditional on the chord length behaves like a gamma distribution with well-behaved parameters, with a mean μ=1.1L and a shape parameter β=L−0.645. Using this information, we are able to generate a transfer function that can adjust the chord length distribution so that it comes much closer to the cloud area distribution. Our transfer function improves the error in predicting the mean cloud size, and is performs without strong biases for smaller sample sizes. However, we find that the method is still has difficulties in accurately predicting the frequency of occurrence of the largest cloud sizes.
Carbon monoxide over the Amazon Basin during the wet season
Measurements of the distribution of carbon monoxide in the lower atmosphere (150-5000 m) over the Amazon region of Brazil during the wet season, taken in conjunction with the 1987 NASA Global Tropospheric Experiment/Amazon Boundary Layer Experiment (ABLE 2B), are analyzed. About 100 hr of airborne, in situ CO measurements were obtained using a tunable diode laser system, providing insights into factors influencing the basin-scale distribution of CO in the Amazonian troposphere during wet season conditions. Distribution of CO over the altitudes 0.15-4.5 km was influenced by such factors as surface emissions from biological sources and long-range transport of pollutants from Northern Hemisphere sources. It is noted that the disruption of mixed layer growth and decay processes has a particularly important influence on CO concentration in the daytime lower troposphere and that the correlation of CO with O3 was positive under conditions influenced by Northern Hemisphere air and negative under all other conditions observed.
Range reference atmosphere models
A description is given of the methods used to establish the statistical parameters and models for wind and various thermodynamic quantities at an altitude of 0-70 km for nine geographical locations. It is noted that wind is modeled as a vector quantity using the bivariate normal probability function. With the five parameters of the bivariate normal distribution, the distribution for wind speed is derived as a generalized Rayleigh distribution. In addition, the frequency of wind direction is derived, and the conditional distribution of wind speed given the wind direction is derived. It is pointed out that these and other wind models are consistent with the rigorous mathematical properties of the bivariate normal probability theory. The thermodynamic quantities are consistent with the hydrostatic equation and the equation of state for the mean values. With these methods, many statistical relationships can be derived.
Calculations of heavy ion charge state distributions for nonequilibrium conditions
Numerical calculations of the charge state distributions of test ions in a hot plasma under nonequilibrium conditions are presented. The mean ionic charges of heavy ions for finite residence times in an instantaneously heated plasma and for a non-Maxwellian electron distribution function are derived. The results are compared with measurements of the charge states of solar energetic particles, and it is found that neither of the two simple cases considered can explain the observations.
NASA GPM GV Science Requirements
An important scientific objective of the NASA portion of the GPM Mission is to generate quantitatively-based error characterization information along with the rainrate retrievals emanating from the GPM constellation of satellites. These data must serve four main purposes: (1) they must be of sufficient quality, uniformity, and timeliness to govern the observation weighting schemes used in the data assimilation modules of numerical weather prediction models; (2) they must extend over that portion of the globe accessible by the GPM core satellite to which the NASA GV program is focused - (approx.65 degree inclination); (3) they must have sufficient specificity to enable detection of physically-formulated microphysical and meteorological weaknesses in the standard physical level 2 rainrate algorithms to be used in the GPM Precipitation Processing System (PPS), i.e., algorithms which will have evolved from the TRMM standard physical level 2 algorithms; and (4) they must support the use of physical error modeling as a primary validation tool and as the eventual replacement of the conventional GV approach of statistically intercomparing surface rainrates fiom ground and satellite measurements. This approach to ground validation research represents a paradigm shift vis-&-vis the program developed for the TRMM mission, which conducted ground validation largely as a statistical intercomparison process between raingauge-derived or radar-derived rainrates and the TRMM satellite rainrate retrievals -- long after the original satellite retrievals were archived. This approach has been able to quantify averaged rainrate differences between the satellite algorithms and the ground instruments, but has not been able to explain causes of algorithm failures or produce error information directly compatible with the cost functions of data assimilation schemes. These schemes require periodic and near-realtime bias uncertainty (i.e., global space-time distributed conditional accuracy of the retrieved rainrates) and local error covariance structure (i.e., global space-time distributed error correlation information for the local 4-dimensional space-time domain -- or in simpler terms, the matrix form of precision error). This can only be accomplished by establishing a network of high quality-heavily instrumented supersites selectively distributed at a few oceanic, continental, and coastal sites. Economics and pragmatics dictate that the network must be made up of a relatively small number of sites (6-8) created through international cooperation. This presentation will address some of the details of the methodology behind the error characterization approach, some proposed solutions for expanding site-developed error properties to regional scales, a data processing and communications concept that would enable rapid implementation of algorithm improvement by the algorithm developers, and the likely available options for developing the supersite network.