arXiv subjects

Andrew Ulmer

Publications and source records attributed to Andrew Ulmer.

13 recordsLinked to original sources

arXiv2022

MABe22: A Multi-Species Multi-Task Benchmark for Learned Representations of Behavior

We introduce MABe22, a large-scale, multi-agent video and trajectory benchmark to assess the quality of learned behavior representations. This dataset is collected from a variety of biology experiments, and includes triplets of interacting mice (4.7 million frames video+pose tracking data, 10 million frames pose only), symbiotic beetle-ant interactions (10 million frames video data), and groups of interacting flies (4.4 million frames of pose tracking data). Accompanying these data, we introduce a panel of real-life downstream analysis tasks to assess the quality of learned representations by evaluating how well they preserve information about the experimental conditions (e.g. strain, time of day, optogenetic stimulation) and animal behavior. We test multiple state-of-the-art self-supervised video and trajectory representation learning methods to demonstrate the use of our benchmark, revealing that methods developed using human action datasets do not fully translate to animal datasets. We hope that our benchmark and dataset encourage a broader exploration of behavior representation learning methods across species and settings.

cs.LG↗

arXiv2019

Cloud Resource Optimization for Processing Multiple Streams of Visual Data

Hundreds of millions of network cameras have been installed throughout the world. Each is capable of providing a vast amount of real-time data. Analyzing the massive data generated by these cameras requires significant computational resources and the demands may vary over time. Cloud computing shows the most promise to provide the needed resources on demand. In this article, we investigate how to allocate cloud resources when analyzing real-time data streams from network cameras. A resource manager considers many factors that affect its decisions, including the types of analysis, the number of data streams, and the locations of the cameras. The manager then selects the most cost-efficient types of cloud instances (e.g. CPU vs. GPGPU) to meet the computational demands for analyzing streams. We evaluate the effectiveness of our approach using Amazon Web Services. Experiments demonstrate more than 50% cost reduction for real workloads.

cs.DC↗

arXiv1997

Tidal Disruption Eddington Envelopes around Massive Black Holes

Optically-thick envelopes may form following the tidal disruption of a star by a massive black hole. Such envelopes would reprocess hard radiation from accretion close to the black hole into the UV and optical bands producing AGN-luminosity flares with duration ~1 year. We show that due to relativistic effects, the envelopes are convective. If convection is efficient, then the structure of the envelopes is similar to that described in previous work; however, the photospheric radius is shown to be very sensitive to the luminosity at the envelope base, suggesting that either the envelope collapses or the envelope expands to a maximum radius at which point a wind may set in. For an envelope without winds, we find a maximum photospheric radius of ~10^16 cm (i.e. minimum effective temperature ~6,000 K). The evolution of the envelopes is described based on simple energy arguments.

astro-ph↗

arXiv1997

Evolution of Thick Accretion Disks Produced by Tidal Disruption Events

Geometrically thick disks may form after tidal disruption events, and rapid accretion may lead to short flares followed by long-term, lower-level emission. Using a novel accretion disk code which relies primarily on global conservation laws and the assumption that viscosity is everywhere positive, a broad range of physically allowed evolutionary sequences of thick disks is investigated. The main result is that accretion in the thick disk phase can consume only a fraction of the initial disk material before the disk cools and becomes thin. This fraction is ~0.5-0.9 for disruptions around 10^6 to 10^7 M_\odot black holes and is sensitive to the mean angular momentum of the disk. The residual material will accrete in some form of thin disk over a longer period of time. The initial thick disk phase may reduce the dimming timescale of the disk by a factor of ~2 from estimates based on thin disks alone. Assuming an 0.5 M_\odot initial thick disk, even if the thin disks become advection dominated, the black hole mass to light ratio can rise above M_\odot/L_\odot = 1 in no less than 20 (0.1/α) to 2000 (0.1/α) years following a tidal disruption event, depending on the mass of the black hole and the initial conditions of the encounter. The long-term emission will be most prevalent around lower mass, 10^6 M_\odot black holes. If the tidal disruption rates in these galactic nuclei are ~10^{-4} yr^{-1}, then about 10% of the nuclei should exhibit the long-term UV/optical emission at a level of ~10^{38} ergs~s^{-1}.

astro-ph↗

arXiv1997

Revisiting the Modified Eddington Limit for Massive Stars

We have determined the location of the line-opacity modified Eddington limit for stars in the LMC using the most recent atmosphere models combined with a precise mapping to the HR Diagram through up-to-date stellar evolution calculations. While we find, in agreement with previous studies, that the shape of the modified Eddington limit qualitatively corresponds to the Humphreys-Davidson (HD) limit defined by the most luminous supergiants, the modified limit is actually a full magnitude higher than the upper luminosity limit observed for LMC stars. The observed limit is consistent with atmosphere models in which the maximum value of the ratio of the radiation force outwards to the gravitational force inwards, Y_max, is 0.9, i.e., the photospheres of stars at the observed luminosity limit are bound. As massive stars evolve, they move to higher, and therefore less stable values of Y_max, so mass loss, either sporadic or continuous, may halt their natural redward evolution as they approach the Y_max = 0.9 limit. We assess the metallicity dependence of this limit. If the limit does determine the most luminous stars, and the value of Y_max corresponding to the luminosity limit in the LMC is universal, then the brightest supergiants the SMC should be only marginally brighter (0.3 mag) than those of the LMC, in agreement with observations. Moreover, the brightest supergiants in M31 should be 0.75~mag fainter than those in the LMC.

astro-ph↗

arXiv1997

Flares from the Tidal Disruption of Stars by Massive Black Holes

Tidal disruption flares are differentiated into two classes -- those which are sub-Eddington and those which radiate near the Eddington limit. Flares from black holes above ~2 x 10^7 M_\odot will generally not radiate above the Eddington limit. For a Schwarzschild black hole, the maximum bolometric luminosity of a tidal disruption is ~L_Edd(5 x 10^7 M_\odot), substantially below the Eddington luminosities of the most massive disrupting black holes (~2 x 10^8 M_\odot). Bolometric corrections to the spectra of the brightest flares are found to be large (~7.5 mag). Nevertheless, the brightest flares are likely to have absolute magnitudes in excess of -19 in V and -21 in U (in the absence of reddening). Because the spectra are so blue, K-corrections may actually brighten the flares in optical bands. If such flares are as frequent as believed, they may soon be detected in low or high redshift supernovae searches. The He II ionizing radiation produced in the flares may dominate that which is produced by all other sources in the centers of quiescent galaxies, creating a steady state, highly ionized, fossil nebula with an extent of ~1 kpc which may be observable in recombination lines.

astro-ph↗

arXiv1997

Optical Appearance of the Debris of a Star Disrupted by a Massive Black Hole

We show that the disruption of a star by a 10^6 solar mass black hole in a galactic nucleus could under favorable circumstances produce an optically-thick envelope that radiates with a thermal spectrum at the Eddington limit, 10^{44} erg/s, for tens of years. The low apparent temperature of this envelope, 10^4 K, would be easily detectable in optical surveys. If most galaxies harbor a massive black hole at their center, then the Sloan Digital Sky Survey might find hundreds of galaxies with nuclear activity of this type. Because the envelope is driven to shine near the Eddington limit, a measurement of the source redshift and total luminosity could yield the black hole mass.

astro-ph↗

arXiv1996

Strong Clustering in the Low Redshift Lyman-$α$ Forest

The two-point correlation function, $ξ$, of Lyman-alpha forest is found to be large, $ξ= 1.8^{+1.6}_{-1.2}$, > 90% confidence level, on the scale of 250-500 km/s for a sample of absorbers (0 < z < 1.3) assembled from HST Key Project Observations. This correlation function is stronger than at high redshift (z > 1.7) where $ξ\approx 0.2$ for velocities > 250 km/s.

astro-ph↗

arXiv1996

The Temperature Dependence of Solar Neutrino Fluxes

By comparing neutrino fluxes and central temperatures calculated from 1000 detailed numerical solar models, we derive improved scaling laws which show how each of the neutrino fluxes depends upon the central temperature (flux $\propto T^m$); we also estimate uncertainties for the temperature exponents. With the aid of a one-zone model of the sun, we derive expressions for the temperature exponents of the neutrino fluxes. For the most important neutrino fluxes, the exponents calculated with the one-zone model agree to within 20\% or better with the exponents extracted from the detailed numerical models. The one-zone model provides a physical understanding of the temperature dependence of the neutrino fluxes. For the $pp$ neutrino flux, the one-zone model explains the (initially-surprising) dependence of the flux upon a negative power of the temperature and suggests a new functional dependence. This new function makes explicit the strong anti-correlation between the $^7$Be and $pp$ neutrino fluxes. The one-zone model also predicts successfully the average linear relations between neutrino fluxes, but cannot predict the appreciable scatter in a $Δϕ_i/ϕ_i$ versus $Δϕ_j/ϕ_j$ diagram.

astro-ph↗

arXiv1994

Constraints on the Gamma-ray Burst Luminosity Function from PVO and BATSE

We examine the width of the gamma-ray burst luminosity function through the distribution of GRB peak fluxes as detected by the Pioneer Venus Orbiter (PVO) and the Burst and Transient Source Experiment (BATSE). The strength of the analysis is greatly enhanced by using a merged catalog of peak fluxes from both instruments with good cross-calibration of their sensitivities. The range of peak fluxes is increased by approximately a factor of 20 relative to the BATSE catalog. Thus, more sensitive investigations of the $\log N-\log P$ distribution are possible. We place constraints on the width of the luminosity function of gamma-ray bursts brighter than the BATSE completeness limit by comparing the intensity distribution in the merged catalog with those produced by a variety of spatial density and luminosity functions. For the models examined, $90\%$ of the {\em detectable\/} bursts have peak luminosities within a range of 10, indicating that the peak luminosities of gamma-ray bursts span a markedly less wide range of values than many other of their measurable properties. We also discuss for which slopes of a power-law luminosity function the observed width is at the upper end of the constrained range. This is important in determining the power-law slopes for which luminosity-duration correlations could be important.

astro-ph↗

arXiv1994

Super-Eddington Radiation Transfer in Soft Gamma Repeaters

Bursts from soft gamma repeaters have been shown to be super-Eddington by a factor of 1000 and have been persuasively associated with compact objects. Here, a model of super-Eddington radiation transfer on the surface of a strongly magnetic ($\geq 10^{13}$ gauss) neutron star is studied and related to the observational constraints on soft gamma repeaters. In strong magnetic fields, the cross-section to electron scattering is strongly suppressed in one polarization state, so super-Eddington fluxes can be radiated while the plasma remains in hydrostatic equilibrium. The model offers a somewhat natural explanation for the observation of similarity between spectra from bursts of varying intensity. The radiation produced in the model is found to be linearly polarized to about 1 part in 1000 in a direction determined by the local magnetic field, and the large intensity variations between bursts are understood as a change in the radiating area on the source. Therefore, the polarization may vary as a function of burst intensity, since the complex structure of the magnetic field may be more apparent for larger radiating areas. It is shown that for radiation transfer calculations in this limit of super-strong magnetic fields it is sufficient to solve the radiation transfer equations for the low opacity state rather than the coupled equations for both. With this approximation, standard stellar atmosphere techniques are utilized to calculate the model energy spectrum.

astro-ph↗

arXiv1994

The Width of the Gamma-ray Burst Luminosity Function

We examine the width of the gamma-ray burst (GRB) luminosity function through the distribution of GRB peak count rates, C$_{\rm{peak}}$, as detected by BATSE (\cite{batse:93}). In the context of galactic corona spatial distribution models, we attempt to place constraints on the characteristic width of the luminosity function by comparing the observed intensity distribution with those produced by a range of density and luminosity functions. We find that the intrinsic width of the luminosity function cannot be very well restricted. However, the distribution of intrinsic luminosities of {\it detected bursts} can be limited: we find that most observed bursts have luminosities that are in a range of one to two decades, but a significant population of undetected less luminous bursts cannot be excluded. These findings demonstrate that the assumption that GRB are standard candles is sufficient but not necessary to explain the observed intensity distribution. We show that the main reason for the relatively poor constraints is the fact that the bright-end part of the GRB flux distribution is not yet sampled by BATSE, and better sampling in the future may lead to significantly stronger constraints on the width of the luminosity function.

astro-ph↗

arXiv1994

Gamma-Rays from Grazing Incidence Cosmic Rays in the Earth's Atmosphere

Interactions of grazing incidence, ultra high energy cosmic rays with the earth's atmosphere may provide a new method of studying energetic cosmic rays with gamma-ray satellites. It is found that these cosmic ray interactions may produce gamma-rays on millisecond time scales which may be detectable by satellites. An extremely low gamma-ray background for transient gamma-ray events and a large area of interaction, the earth's surface, make the scheme plausible. The effective cross section of detection of interactions for cosmic rays above 10^20 eV is found to be more than two orders of magnitude higher than earth based detection techniques. This method may eventually offer an efficient way of probing this region of the cosmic ray energy spectrum where events are scarce. In this paper, a conceptual model is presented for the production of short bursts of gamma-rays based on these grazing incidence encounters with the earth's atmosphere.

astro-ph↗