skip to main content


Title: The isoperimetric problem in the plane with the sum of two Gaussian densities
We consider the isoperimetric problem for the sum of two Gaussian densities in the line and the plane. We prove that the double Gaussian isoperimetric regions in the line are rays and that if the double Gaussian isoperimetric regions in the plane are half-spaces, then they must be bounded by vertical lines.  more » « less
Award ID(s):
1561945 1659037
NSF-PAR ID:
10054850
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Involve
Volume:
11
Issue:
4
ISSN:
1944-4184
Page Range / eLocation ID:
549-567
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Abstract We investigate the kinematic properties of Galactic H ii regions using radio recombination line (RRL) emission detected by the Australia Telescope Compact Array at 4–10 GHz and the Jansky Very Large Array at 8–10 GHz. Our H ii region sample consists of 425 independent observations of 374 nebulae that are relatively well isolated from other, potentially confusing sources and have a single RRL component with a high signal-to-noise ratio. We perform Gaussian fits to the RRL emission in position-position–velocity data cubes and discover velocity gradients in 178 (42%) of the nebulae with magnitudes between 5 and 200 m s − 1 arcsec − 1 . About 15% of the sources also have an RRL width spatial distribution that peaks toward the center of the nebula. The velocity gradient position angles appear to be random on the sky with no favored orientation with respect to the Galactic plane. We craft H ii region simulations that include bipolar outflows or solid body rotational motions to explain the observed velocity gradients. The simulations favor solid body rotation since, unlike the bipolar outflow kinematic models, they are able to produce both the large, >40 m s − 1 arcsec − 1 , velocity gradients and also the RRL width structure that we observe in some sources. The bipolar outflow model, however, cannot be ruled out as a possible explanation for the observed velocity gradients for many sources in our sample. We nevertheless suggest that most H ii region complexes are rotating and may have inherited angular momentum from their parent molecular clouds. 
    more » « less
  2. Deep generative models parametrized up to a normalizing constant (e.g. energy-based models) are difficult to train by maximizing the likelihood of the data because the likelihood and/or gradients thereof cannot be explicitly or efficiently written down. Score matching is a training method, whereby instead of fitting the likelihood for the training data, we instead fit the score function, obviating the need to evaluate the partition function. Though this estimator is known to be consistent, it's unclear whether (and when) its statistical efficiency is comparable to that of maximum likelihood---which is known to be (asymptotically) optimal. We initiate this line of inquiry in this paper and show a tight connection between statistical efficiency of score matching and the isoperimetric properties of the distribution being estimated---i.e. the Poincar\'e, log-Sobolev and isoperimetric constant---quantities which govern the mixing time of Markov processes like Langevin dynamics. Roughly, we show that the score matching estimator is statistically comparable to the maximum likelihood when the distribution has a small isoperimetric constant. Conversely, if the distribution has a large isoperimetric constant---even for simple families of distributions like exponential families with rich enough sufficient statistics---score matching will be substantially less efficient than maximum likelihood. We suitably formalize these results both in the finite sample regime, and in the asymptotic regime. Finally, we identify a direct parallel in the discrete setting, where we connect the statistical properties of pseudolikelihood estimation with approximate tensorization of entropy and the Glauber dynamics. 
    more » « less
  3. Optical microscopy suffers from multiple scattering (MS), which limits the optical imaging depth into scattering media. We previously demonstrated aberration-diverse optical coherence tomography (AD-OCT) for MS suppression, based on the principle that for datasets acquired with different aberration states of the imaging beam, MS backgrounds become decorrelated while single scattering (SS) signals remain correlated, so that a simple coherent average can be used to enhance the SS signal over the MS background. Here, we propose a space/spatial-frequency domain analysis framework for the investigation of MS in OCT, and apply the framework to compare AD-OCT (using astigmatic beams) to standard Gaussian-beam OCT via experiments in scattering tissue phantoms. Utilizing this framework, we found that increasing the astigmatic magnitude produced a large drop in both MS background and SS signal, but the decay experienced by the MS background was larger than the SS signal. Accounting for the decay in both SS signal and MS background, the overall signal-to-background ratio (SBR) of AD-OCT was similar to the Gaussian control after about 10 coherent averages, when deeper line foci was positioned at the plane-of-interest and the line foci spacing was smaller than or equal to 80 µm. For an even larger line foci spacing of 160 µm, AD-OCT resulted in a lower SBR than the Gaussian-beam control. This work provides an analysis framework to gain deeper levels of understanding and insights for the future study of MS and MS suppression in both the space and spatial-frequency domains.

     
    more » « less
  4. Abstract

    Turning while walking is ubiquitous and requires linear and angular momenta generation to redirect the body’s trajectory and rotate towards the new direction of travel. This study examined strategies that healthy young adults used during each gait phase to generate transverse-plane momenta during pre-planned and late-cued 90° turns. During leftward turns, we expected that momenta would be generated most during the gait phases known to generate leftward linear and angular momenta during straight line gait. We found distinct roles of gait phases towards generating momenta during turns that partially supported our hypotheses. Supporting one hypothesis, the change in transverse-plane angular momentum and average moment were greater during double support with the left foot in front vs. other gait phases. Also, the change in leftward linear momentum and average leftward force were greater during right single support vs. other gait phases during straight-line gait and late-cued turns. However, during pre-planned turns, the average leftward force was not significantly greater during right single support vs. other gait phases. Overall, transverse-plane angular momentum generation during turns is similar to its generation during straight-line gait, suggesting that healthy young adults can leverage momenta control strategies used during straight-line gait during turns.

     
    more » « less
  5. Concentration of measure has been argued to be the fundamental cause of adversarial vulnerability. Mahloujifar et al. (2019) presented an empirical way to measure the concentration of a data distribution using samples, and employed it to find lower bounds on intrinsic robustness for several benchmark datasets. However, it remains unclear whether these lower bounds are tight enough to provide a useful approximation for the intrinsic robustness of a dataset. To gain a deeper understanding of the concentration of measure phenomenon, we first extend the Gaussian Isoperimetric Inequality to non-spherical Gaussian measures and arbitrary ℓp-norms (p ≥ 2). We leverage these theoretical insights to design a method that uses half-spaces to estimate the concentration of any empirical dataset under ℓp-norm distance metrics. Our proposed algorithm is more efficient than Mahloujifar et al. (2019)‘s, and experiments on synthetic datasets and image benchmarks demonstrate that it is able to find much tighter intrinsic robustness bounds. These tighter estimates provide further evidence that rules out intrinsic dataset concentration as a possible explanation for the adversarial vulnerability of state-of-the-art classifiers. 
    more » « less