Search NASA⌕ Search

Engineering topics

Schneider, Michael D.

Publications and source records attributed to Schneider, Michael D..

Exploration with Scalable Gaussian Process Reinforcement Learning

Exploration is a challenging problem in reinforcement learning (RL), especially in environments with sparse rewards. Quantifying and utilizing the parametric uncertainty has been shown to be paramount for successful exploration [Osband et al., 2018]. Bayesian, or approximately Bayesian, methods present a principled means of estimating the parametric uncertainty in RL problems. Gaussian processes, nonparametric Bayesian models, are often impractical due to poor scalability and computational bottlenecks. We introduce a scalable Gaussian process RL (GPRL) method which directly induces sparsity in the covariance matrix to facilitate faster computation. This is a departure from previous GPRL methods which instead rely on data reduction and subsampling. We compare various covariance-based exploration techniques (Thompson sampling, upper confidence bound, and probabilistic maximum variance) which leverage our scalable GP framework in sparse reward environments. Finally, we show favorable comparison against the bootstrapped deep Q-Network.

97 MATHEMATICS AND COMPUTING↗

A Reanalysis of Public Galactic Bulge Gravitational Microlensing Events from OGLE-III and -IV

Abstract Modern surveys of gravitational microlensing events have progressed to detecting thousands per year, and surveys are capable of probing Galactic structure, stellar evolution, lens populations, black hole physics, and the nature of dark matter. One of the key avenues for doing this is the microlensing Einstein radius crossing time (t E) distribution. However, systematics in individual light curves as well as oversimplistic modeling can lead to biased results. To address this, we developed a model to simultaneously handle the microlensing parallax due to Earth's motion, systematic instrumental effects, and unlensed stellar variability with a Gaussian process model. We used light curves for nearly 10,000 OGLE-III and -IV Milky Way bulge microlensing events and fit each with our model. We also developed a forward model approach to infer the t E distribution by forward modeling from the data rather than using point estimates from individual events. We find that modeling the variability in the baseline removes a source of significant bias in individual events, and the previous analyses overestimated the number of t E > 100 day events due to their oversimplistic model ignoring parallax effects. We use our fits to identify the hundreds filling a regime in the microlensing parameter space that are 50% pure of black holes. Finally, we have released the largest-ever catalog of Markov Chain Monte Carlo parameter estimates for microlensing events.

79 ASTRONOMY AND ASTROPHYSICS↗

The impact of tomographic redshift bin width errors on cosmological probes

ABSTRACT Systematic errors in the galaxy redshift distribution n(z) can propagate to systematic errors in the derived cosmology. We characterize how the degenerate effects in tomographic bin widths and galaxy bias impart systematic errors on cosmology inference using observational data from the Deep Lens Survey. For this we use a combination of galaxy clustering and galaxy–galaxy lensing. We present two end-to-end analyses from the catalogue level to parameter estimation. We produce an initial cosmological inference using fiducial tomographic redshift bins derived from photometric redshifts, then compare this with a result where the redshift bins are empirically corrected using a set of spectroscopic redshifts. We find that the derived parameter S8 ≡ σ8(Ωm/.3)1/2 decreases from $0.841^{+0.062}_{-0.061}$ to $0.781^{0.061}_{0.054}$ upon correcting the n(z) errors in the second method.

79 ASTRONOMY AND ASTROPHYSICS↗

Rare Events via Cross-Entropy Population Monte Carlo

Rare events are events that happen with very low frequency. Estimating rare event probabilities using Monte Carlo techniques is computationally expensive, often to the point of intractability, and special methods are required. Importance sampling (IS) is a well known technique that uses a proposal distribution in place of a target distribution to lower the variance of the estimator. Key to the success of IS methods is the choice of a proposal distribution, or the parameters governing the distribution. Adaptive importance sampling improves the parameters of a family or population of proposal distributions iteratively through trials. We present a novel cross-entropy population Monte Carlo algorithm, which adapts the parameters of proposals through the cross-entropy method. The proposed method stands apart from previous work in that we are not optimizing a mixture distribution. Instead, we leverage deterministic mixture weights and optimize the distributions individually through a reinterpretation of the typical derivation of the cross-entropy method. Demonstrations on rare event examples show that the algorithm can outperform existing resampling based population Monte Carlo methods, especially for higher-dimensional problems. Finally, we also demonstrate efficacy on a conjunction analysis problem.

97 MATHEMATICS AND COMPUTING↗

Reinforcement Learning via Gaussian Processes with Neural Network Dual Kernels

While deep neural networks (DNNs) and Gaussian Processes (GPs) are both popularly utilized to solve problems in reinforcement learning, both approaches feature undesirable drawbacks for challenging problems. DNNs learn complex non-linear embeddings, but do not naturally quantify uncertainty and are often data-inefficient to train. GPs infer posterior distributions over functions, but popular kernels exhibit limited expressivity on complex and high-dimensional data. Fortunately, recently discovered conjugate and neural tangent kernel functions encode the behavior of overparameterized neural networks in the kernel domain. We demonstrate that these kernels can be efficiently applied to regression and reinforcement learning problems by analyzing a baseline case study.We apply GPs with neural network dual kernels to solve reinforcement learning tasks for the first time. We demonstrate, using the well understood mountain-car problem, that GPs empowered with dual kernels perform at least as well as those using the conventional radial basis function kernel. Finally, we conjecture that by inheriting the probabilistic rigor of GPs and the powerful embedding properties of DNNs, GPs using NN dual kernels will empower future reinforcement learning models on difficult domains.

97 MATHEMATICS AND COMPUTING↗

Bayesian Model Selection as a Continuous-Variable Helmholtz Machine

We show that Bayesian model selection is equivalent to optimization of a suitably defined free energy. The entropy term in the free energy is defined over the space of possible model selections and model parameters for a given model. Bayesian model selection thus follows a minimum description length principle.

97 MATHEMATICS AND COMPUTING↗

Cosmic shear results from the deep lens survey. II. Full cosmological parameter constraints from tomography

We present a tomographic cosmic shear study from the Deep Lens Survey (DLS), which, providing a limiting magnitude r{sub lim}∼27 (5σ), is designed as a precursor Large Synoptic Survey Telescope (LSST) survey with an emphasis on depth. Using five tomographic redshift bins, we study their auto- and cross-correlations to constrain cosmological parameters. We use a luminosity-dependent nonlinear model to account for the astrophysical systematics originating from intrinsic alignments of galaxy shapes. We find that the cosmological leverage of the DLS is among the highest among existing >10 deg{sup 2} cosmic shear surveys. Combining the DLS tomography with the 9 yr results of the Wilkinson Microwave Anisotropy Probe (WMAP9) gives Ω{sub m}=0.293{sub −0.014}{sup +0.012}, σ{sub 8}=0.833{sub −0.018}{sup +0.011}, H{sub 0}=68.6{sub −1.2}{sup +1.4} km s{sup −1} Mpc{sup −1}, and Ω{sub b}=0.0475±0.0012 for ΛCDM, reducing the uncertainties of the WMAP9-only constraints by ∼50%. When we do not assume flatness for ΛCDM, we obtain the curvature constraint Ω{sub k}=−0.010{sub −0.015}{sup +0.013} from the DLS+WMAP9 combination, which, however, is not well constrained when WMAP9 is used alone. The dark energy equation-of-state parameter w is tightly constrained when baryonic acoustic oscillation (BAO) data are added, yielding w=−1.02{sub −0.09}{sup +0.10} with the DLS+WMAP9+BAO joint probe. The addition of supernova constraints further tightens the parameter to w=−1.03±0.03. Our joint constraints are fully consistent with the final Planck results and also with the predictions of a ΛCDM universe.

79 ASTRONOMY AND ASTROPHYSICS↗

The ellipticity distribution of ambiguously blended objects

Using overlapping fields with space-based Hubble Space Telescope and ground-based Subaru Telescope imaging we identify a population of blended galaxies that are blended to such a large degree that they are detected as single objects in the ground-based monochromatic imaging, which we label “ambiguous blends.” For deep imaging data, such as the depth targeted with the Large Synoptic Survey Telescope (LSST), the ambiguous blend population is both large (∼14%) and has a distribution of ellipticities that is different from that of unblended objects in a way that will likely be important for weak lensing measurements. Most notably, for a limiting magnitude of i ∼ 27 we find that ambiguous blending results in a ∼14% increase in shear noise (or an ∼12% decrease in the effective projected number density of lensed galaxies; n{sub eff}) due to (1) larger intrinsic ellipticity dispersion, and (2) a scaling with the galaxy number density N{sub gal} that is shallower than 1/√(N{sub gal}). For the LSST Gold Sample (i < 25.3) there is a ∼7% increase in shear noise (or ∼7% decrease in n{sub eff}). More importantly than these increases in the shear noise, we find that the ellipticity distribution of ambiguous blends has an rms that is 13% larger than that of non-blended galaxies. Given the need of future weak lensing surveys to constrain the ellipticity distribution of galaxies to better than a percent in order to mitigate cosmic shear multiplicative biases, if it is unaccounted for, the different ellipticity distribution of ambiguous blends could be a dominant systematic.

79 ASTRONOMY AND ASTROPHYSICS↗

HIERARCHICAL PROBABILISTIC INFERENCE OF COSMIC SHEAR

Point estimators for the shearing of galaxy images induced by gravitational lensing involve a complex inverse problem in the presence of noise, pixelization, and model uncertainties. We present a probabilistic forward modeling approach to gravitational lensing inference that has the potential to mitigate the biased inferences in most common point estimators and is practical for upcoming lensing surveys. The first part of our statistical framework requires specification of a likelihood function for the pixel data in an imaging survey given parameterized models for the galaxies in the images. We derive the lensing shear posterior by marginalizing over all intrinsic galaxy properties that contribute to the pixel data (i.e., not limited to galaxy ellipticities) and learn the distributions for the intrinsic galaxy properties via hierarchical inference with a suitably flexible conditional probabilitiy distribution specification. We use importance sampling to separate the modeling of small imaging areas from the global shear inference, thereby rendering our algorithm computationally tractable for large surveys. With simple numerical examples we demonstrate the improvements in accuracy from our importance sampling approach, as well as the significance of the conditional distribution specification for the intrinsic galaxy properties when the data are generated from an unknown number of distinct galaxy populations with different morphological characteristics.

79 ASTRONOMY AND ASTROPHYSICS↗

GREAT3 results - I. Systematic errors in shear estimation and the impact of real galaxy morphology

The study present first results from the third GRavitational lEnsing Accuracy Testing (GREAT3) challenge, the third in a sequence of challenges for testing methods of inferring weak gravitational lensing shear distortions from simulated galaxy images. GREAT3 was divided into experiments to test three specific questions, and included simulated space- and ground-based data with constant or cosmologically varying shear fields. The simplest (control) experiment included parametric galaxies with a realistic distribution of signal-to-noise, size, and ellipticity, and a complex point spread function (PSF). The other experiments tested the additional impact of realistic galaxy morphology, multiple exposure imaging, and the uncertainty about a spatially varying PSF; the last two questions will be explored in Paper II. The 24 participating teams competed to estimate lensing shears to within systematic error tolerances for upcoming Stage-IV dark energy surveys, making 1525 submissions overall. GREAT3 saw considerable variety and innovation in the types of methods applied. Several teams now meet or exceed the targets in many of the tests conducted (to within the statistical errors). We conclude that the presence of realistic galaxy morphology in simulations changes shear calibration biases by ~1 per cent for a wide range of methods. Other effects such as truncation biases due to finite galaxy postage stamps, and the impact of galaxy type as measured by the Sérsic index, are quantified for the first time. Our results generalize previous studies regarding sensitivities to galaxy size and signal-to-noise, and to PSF properties such as seeing and defocus. Almost all methods’ results support the simple model in which additive shear biases depend linearly on PSF ellipticity.

79 ASTRONOMY AND ASTROPHYSICS↗

COSMIC SHEAR RESULTS FROM THE DEEP LENS SURVEY. I. JOINT CONSTRAINTS ON {Omega} {sub M} AND {sigma}{sub 8} WITH A TWO-DIMENSIONAL ANALYSIS

We present a cosmic shear study from the Deep Lens Survey (DLS), a deep BVRz multi-band imaging survey of five 4 deg{sup 2} fields with two National Optical Astronomy Observatory (NOAO) 4 m telescopes at Kitt Peak and Cerro Tololo. For both telescopes, the change of the point-spread-function (PSF) shape across the focal plane is complicated, and the exposure-to-exposure variation of this position-dependent PSF change is significant. We overcome this challenge by modeling the PSF separately for individual exposures and CCDs with principal component analysis (PCA). We find that stacking these PSFs reproduces the final PSF pattern on the mosaic image with high fidelity, and the method successfully separates PSF-induced systematics from gravitational lensing effects. We calibrate our shears and estimate the errors, utilizing an image simulator, which generates sheared ground-based galaxy images from deep Hubble Space Telescope archival data with a realistic atmospheric turbulence model. For cosmological parameter constraints, we marginalize over shear calibration error, photometric redshift uncertainty, and the Hubble constant. We use cosmology-dependent covariances for the Markov Chain Monte Carlo analysis and find that the role of this varying covariance is critical in our parameter estimation. Our current non-tomographic analysis alone constrains the {Omega} {sub M}-{sigma}{sub 8} likelihood contour tightly, providing a joint constraint of {Omega} {sub M} = 0.262 {+-} 0.051 and {sigma}{sub 8} = 0.868 {+-} 0.071. We expect that a future DLS weak-lensing tomographic study will further tighten these constraints because explicit treatment of the redshift dependence of cosmic shear more efficiently breaks the {Omega} {sub M}-{sigma}{sub 8} degeneracy. Combining the current results with the Wilkinson Microwave Anisotropy Probe 7 year (WMAP7) likelihood data, we obtain {Omega} {sub M} = 0.278 {+-} 0.018 and {sigma}{sub 8} = 0.815 {+-} 0.020.

79 ASTRONOMY AND ASTROPHYSICS↗

The shapes and alignments of dark matter halos

We present measurements of the triaxial dark matter halo shapes and alignment correlation functions in the Millennium and Millennium-2 dark matter N-body simulations. These two simulations allow us to measure the distributions of halo shapes down to 10% of the virial radius over a halo mass range of 6 × 10{sup 9}–2 × 10{sup 14} h{sup −1}M{sub s}un. We largely confirm previous results on the distributions of halo axis ratios as a function of halo mass, but we find that the median angle between halo major axes at different halo radii can vary by a factor of 2 between the Millennium-1 and 2 simulations because of the different mass resolution. Thus, error in the shape determinations from limited resolution is potentially degenerate with the misalignment of halo inner and outer shapes used to constrain Brightest Cluster Galaxy alignments in previous works. We also present simplifying parameterizations for the 3-D halo-mass alignment correlation functions that are necessary ingredients for triaxial halo models of large-scale structure and models of galaxy intrinsic alignments as contaminants for cosmic shear surveys. We measure strong alignments between halos of all masses and the surrounding dark matter overdensities out to several tens of h{sup −1} Mpc, in agreement with observed shear-galaxy and cluster shape correlations. We use these measurements to forecast the contribution to the weak lensing signal around galaxy clusters from correlated mass along the line-of-sight. For prolate clusters with major axes aligned with the line-of-sight the fraction of the weak lensing signal from mass external to the cluster can be twice that predicted if the excess halo alignment correlation is assumed to be zero.

79 ASTRONOMY AND ASTROPHYSICS↗

Contractile reserve and calcium regulation are depressed in myocytes from chronically unloaded hearts

BACKGROUND: Chronic cardiac unloading of the normal heart results in the reduction of left ventricular (LV) mass, but effects on myocyte contractile function are not known. METHODS AND RESULTS: Cardiac unloading and reduction in LV mass were induced by heterotopic heart transplantation to the abdominal aorta in isogenic rats. Contractility and [Ca(2+)](i) regulation in LV myocytes were studied at both 2 and 5 weeks after transplantation. Native in situ hearts from recipient animals were used as the controls for all experiments. Contractile function indices in myocytes from 2-week unloaded and native (control) hearts were similar under baseline conditions (0.5 Hz, 1.2 mmol/L [Ca(2+)](o), and 36 degrees C) and in response to stimulation with high [Ca(2+)](o) (range 2.5 to 4.0 mmol/L). In myocytes from 5-week unloaded hearts, there were no differences in fractional cell shortening and peak-systolic [Ca(2+)](i) at baseline; however, time to 50% relengthening and time to 50% decline in [Ca(2+)](i) were prolonged compared with controls. Severe defects in fractional cell shortening and peak-systolic [Ca(2+)](i) were elicited in myocytes from 5-week unloaded hearts in response to high [Ca(2+)](o). However, there were no differences in the contractile response to isoproterenol between myocytes from unloaded and native hearts. In 5-week unloaded hearts, but not in 2-week unloaded hearts, LV protein levels of phospholamban were increased (345% of native heart values). Protein levels of sarcoplasmic reticulum Ca(2+) ATPase and the Na(+)/Ca(2+) exchanger were not changed. CONCLUSIONS: Chronic unloading of the normal heart caused a time-dependent depression of myocyte contractile function, suggesting the potential for impaired performance in states associated with prolonged cardiac atrophy.

NASA Discipline Cardiopulmonary↗