arXiv subjects

Andrew Howard

Publications and source records attributed to Andrew Howard.

At least 55 records · Page 3Linked to original sources

arXiv2018

2018 Low-Power Image Recognition Challenge

The Low-Power Image Recognition Challenge (LPIRC, https://rebootingcomputing.ieee.org/lpirc) is an annual competition started in 2015. The competition identifies the best technologies that can classify and detect objects in images efficiently (short execution time and low energy consumption) and accurately (high precision). Over the four years, the winners' scores have improved more than 24 times. As computer vision is widely used in many battery-powered systems (such as drones and mobile phones), the need for low-power computer vision will become increasingly important. This paper summarizes LPIRC 2018 by describing the three different tracks and the winners' solutions.

cs.CV↗

arXiv2018

NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications

This work proposes an algorithm, called NetAdapt, that automatically adapts a pre-trained deep neural network to a mobile platform given a resource budget. While many existing algorithms simplify networks based on the number of MACs or weights, optimizing those indirect metrics may not necessarily reduce the direct metrics, such as latency and energy consumption. To solve this problem, NetAdapt incorporates direct metrics into its adaptation algorithm. These direct metrics are evaluated using empirical measurements, so that detailed knowledge of the platform and toolchain is not required. NetAdapt automatically and progressively simplifies a pre-trained network until the resource budget is met while maximizing the accuracy. Experiment results show that NetAdapt achieves better accuracy versus latency trade-offs on both mobile CPU and mobile GPU, compared with the state-of-the-art automated network simplification algorithms. For image classification on the ImageNet dataset, NetAdapt achieves up to a 1.7$\times$ speedup in measured inference latency with equal or higher accuracy on MobileNets (V1&V2).

cs.CV↗

arXiv2018

Direct Imaging in Reflected Light: Characterization of Older, Temperate Exoplanets With 30-m Telescopes

Direct detection, also known as direct imaging, is a method for discovering and characterizing the atmospheres of planets at intermediate and wide separations. It is the only means of obtaining spectra of non-transiting exoplanets. Characterizing the atmospheres of planets in the <5 AU regime, where RV surveys have revealed an abundance of other worlds, requires a 30-m-class aperture in combination with an advanced adaptive optics system, coronagraph, and suite of spectrometers and imagers - this concept underlies planned instruments for both TMT (the Planetary Systems Imager, or PSI) and the GMT (GMagAO-X). These instruments could provide astrometry, photometry, and spectroscopy of an unprecedented sample of rocky planets, ice giants, and gas giants. For the first time habitable zone exoplanets will become accessible to direct imaging, and these instruments have the potential to detect and characterize the innermost regions of nearby M-dwarf planetary systems in reflected light. High-resolution spectroscopy will not only illuminate the physics and chemistry of exo-atmospheres, but may also probe rocky, temperate worlds for signs of life in the form of atmospheric biomarkers (combinations of water, oxygen and other molecular species). By completing the census of non-transiting worlds at a range of separations from their host stars, these instruments will provide the final pieces to the puzzle of planetary demographics. This whitepaper explores the science goals of direct imaging on 30-m telescopes and the technology development needed to achieve them.

astro-ph.IM↗

arXiv2018

The TRENDS High-Contrast Imaging Survey. VII. Discovery of a Nearby Sirius-like White Dwarf System (HD 169889)

Monitoring the long-term radial velocity (RV) and acceleration of nearby stars has proven an effective method for directly detecting binary and substellar companions. Some fraction of nearby RV trend systems are expected to be comprised of compact objects that likewise induce a systemic Doppler signal. In this paper, we report the discovery of a white dwarf companion found to orbit the nearby ($π= 28.297 \pm 0.066$ mas) G9 V star HD 169889. High-contrast imaging observations using NIRC2 at Keck and LMIRCam at the LBT uncover the ($ΔH = 9.76 \pm 0.16$, $ΔL' = 9.60 \pm 0.03$) companion at an angular separation of 0.8'' (28 au). Thirteen years of precise Doppler observations reveal a steep linear acceleration in RV time series and place a dynamical constraint on the companion mass of $M \geq 0.369 \pm 0.010 M_{\odot}$. This "Sirius-like" system adds to the census of white dwarf companions suspected to be missing in the solar neighborhood.

astro-ph.SR↗

arXiv2018

Large Scale Fine-Grained Categorization and Domain-Specific Transfer Learning

Transferring the knowledge learned from large scale datasets (e.g., ImageNet) via fine-tuning offers an effective solution for domain-specific fine-grained visual categorization (FGVC) tasks (e.g., recognizing bird species or car make and model). In such scenarios, data annotation often calls for specialized domain knowledge and thus is difficult to scale. In this work, we first tackle a problem in large scale FGVC. Our method won first place in iNaturalist 2017 large scale species classification challenge. Central to the success of our approach is a training scheme that uses higher image resolution and deals with the long-tailed distribution of training data. Next, we study transfer learning via fine-tuning from large scale datasets to small scale, domain-specific FGVC datasets. We propose a measure to estimate domain similarity via Earth Mover's Distance and demonstrate that transfer learning benefits from pre-training on a source domain that is similar to the target domain by this measure. Our proposed transfer learning outperforms ImageNet pre-training and obtains state-of-the-art results on multiple commonly used FGVC datasets.

cs.CV↗

arXiv2018

EarthFinder: A Precise Radial Velocity Probe Mission Concept For the Detection of Earth-Mass Planets Orbiting Sun-like Stars

EarthFinder is a Probe Mission concept selected for study by NASA for input to the 2020 astronomy decadal survey. This study is currently active and a final white paper report is due to NASA at the end of calendar 2018. We are tasked with evaluating the scientific rationale for obtaining precise radial velocity (PRV) measurements in space, which is a two-part inquiry: What can be gained from going to space? What can't be done form the ground? These two questions flow down to these specific tasks for our study - Identify the velocity limit, if any, introduced from micro- and macro-telluric absorption in the Earth's atmosphere; Evaluate the unique advantages that a space-based platform provides to emable the identification and mitigation of stellar acitivity for multi-planet signal recovery.

astro-ph.IM↗

arXiv2018

HD 4915: A Maunder Minimum Candidate

We study the magnetic activity cycle of HD 4915 using the \ion{Ca}{2} H \& K emission line strengths measured by Keck I/HIRES spectrograph. The star has been observed as a part of California Planet Search Program from 2006 to present. We note decreasing amplitude in the magnetic activity cycle, a pattern suggesting the star's entry into a Magnetic Grand Minimum (MGM) state, reminiscent of the Sun's Maunder and Dalton Minima. We recommend further monitoring of the star to confirm the grand minimum nature of the dynamo, which would provide insight into the state of the Sun's chromosphere and the global magnetic field during its grand minima. We also recommend continued observations of H \& K emission lines, and ground or space based photometric observations to estimate the sunspot coverage.

astro-ph.SR↗

arXiv2017

Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference

The rising popularity of intelligent mobile devices and the daunting computational cost of deep learning-based models call for efficient and accurate on-device inference schemes. We propose a quantization scheme that allows inference to be carried out using integer-only arithmetic, which can be implemented more efficiently than floating point inference on commonly available integer-only hardware. We also co-design a training procedure to preserve end-to-end model accuracy post quantization. As a result, the proposed quantization scheme improves the tradeoff between accuracy and on-device latency. The improvements are significant even on MobileNets, a model family known for run-time efficiency, and are demonstrated in ImageNet classification and COCO detection on popular CPUs.

cs.LG↗

arXiv2017

Seeing double with K2: Testing re-inflation with two remarkably similar planets around red giant branch stars

Despite more than 20 years since the discovery of the first gas giant planet with an anomalously large radius, the mechanism for planet inflation remains unknown. Here, we report the discovery of EPIC228754001.01, an inflated gas giant planet found with the NASA K2 Mission, and a revised mass for another inflated planet, K2-97b. These planets reside on ~9 day orbits around host stars which recently evolved into red giants. We constrain the irradiation history of these planets using models constrained by asteroseismology and Keck/HIRES spectroscopy and radial velocity measurements. We measure planet radii of 1.31 +\- 0.11 Rjup and and 1.30 +\- 0.07 Rjup, respectively. These radii are typical for planets receiving the current irradiation, but not the former, zero age main sequence irradiation of these planets. This suggests that the current sizes of these planets are directly correlated to their current irradiation. Our precise constraints of the masses and radii of the stars and planets in these systems allow us to constrain the planetary heating efficiency of both systems as 0.03% +0.03%/-0.02%. These results are consistent with a planet re-inflation scenario, but suggest the efficiency of planet re-inflation may be lower than previously theorized. Finally, we discuss the agreement within 10% of stellar masses and radii, and planet masses, radii, and orbital periods of both systems and speculate that this may be due to selection bias in searching for planets around evolved stars.

astro-ph.EP↗

arXiv2016

The Unreasonable Effectiveness of Noisy Data for Fine-Grained Recognition

Current approaches for fine-grained recognition do the following: First, recruit experts to annotate a dataset of images, optionally also collecting more structured data in the form of part annotations and bounding boxes. Second, train a model utilizing this data. Toward the goal of solving fine-grained recognition, we introduce an alternative approach, leveraging free, noisy data from the web and simple, generic methods of recognition. This approach has benefits in both performance and scalability. We demonstrate its efficacy on four fine-grained datasets, greatly exceeding existing state of the art without the manual collection of even a single label, and furthermore show first results at scaling to more than 10,000 fine-grained categories. Quantitatively, we achieve top-1 accuracies of 92.3% on CUB-200-2011, 85.4% on Birdsnap, 93.4% on FGVC-Aircraft, and 80.8% on Stanford Dogs without using their annotated training sets. We compare our approach to an active learning approach for expanding fine-grained datasets.

cs.CV↗

arXiv2016

Spitzer Observations of Exoplanets Discovered with The Kepler K2 Mission

We have used the {\it Spitzer Space Telescope} to observe two transiting planetary systems orbiting low mass stars discovered in the \Kepler \Ktwo mission. The system K2-3 (EPIC 201367065) hosts three planets while EPIC 202083828 (K2-26) hosts a single planet. Observations of all four objects in these two systems confirm and refine the orbital and physical parameters of the planets. The refined orbital information and more precise planet radii possible with \Spitzer will be critical for future observations of these and other \Ktwo targets. For K2-3b we find marginally significant evidence for a Transit Timing Variation between the \Ktwo and \Spitzer\ epochs.

astro-ph.EP↗

arXiv2015

Two Stars Two Ways: Confirming a Microlensing Binary Lens Solution with a Spectroscopic Measurement of the Orbit

Light curves of microlensing events involving stellar binaries and planetary systems can provide information about the orbital elements of the system due to orbital modulations of the caustic structure. Accurately measuring the orbit in either the stellar or planetary case requires detailed modeling of subtle deviations in the light curve. At the same time, the natural, Cartesian parameterization of a microlensing binary is partially degenerate with the microlens parallax. Hence, it is desirable to perform independent tests of the predictions of microlens orbit models using radial velocity time series of the lens binary system. To this end, we present 3.5 years of RV monitoring of the binary lens system OGLE-2009-BLG-020L, for which Skowron et al. (2011) constrained all internal parameters of the 200--700 day orbit. Our RV measurements reveal an orbit that is consistent with the predictions of the microlens light curve analysis, thereby providing the first confirmation of orbital elements inferred from microlensing events.

astro-ph.SR↗

arXiv2014

Empirical Limits on Radial Velocity Planet Detection for Young Stars

We report initial results from our long term search using precision radial velocities for planetary-mass companions located within a few AU of stars younger than the Sun. Based on a sample of >150 stars, we define a floor in the radial velocity scatter, sigma_RV, as a function of the chromospheric activity level R'_{HK}. This lower bound to the jitter, which increases with increasing stellar activity, sets the minimum planet mass that could be detected. Adopting a median activity-age relationship reveals the astrophysical limits to planet masses discernable via radial velocity monitoring, as a function of stellar age. Considering solar-mass primaries having the mean jitter-activity level, when they are younger than 100 / 300 / 1000 Myr, the stochastic jitter component in radial velocity measurements restricts detectable companion masses to > 0.3 / 0.2 / 0.1 M_Jupiter. These numbers require a large number -- several tens -- of radial velocity observations taken over a time frame longer than the orbital period. Lower companion mass limits can be achieved for stars with less than the mean jitter and/or with an increased number of observations.

astro-ph.SR↗

arXiv2012

Dynamical Systems Trees

We propose dynamical systems trees (DSTs) as a flexible class of models for describing multiple processes that interact via a hierarchy of aggregating parent chains. DSTs extend Kalman filters, hidden Markov models and nonlinear dynamical systems to an interactive group scenario. Various individual processes interact as communities and sub-communities in a tree structure that is unrolled in time. To accommodate nonlinear temporal activity, each individual leaf process is modeled as a dynamical system containing discrete and/or continuous hidden states with discrete and/or Gaussian emissions. Subsequent higher level parent processes act like hidden Markov models and mediate the interaction between leaf processes or between other parent processes in the hierarchy. Aggregator chains are parents of child processes that they combine and mediate, yielding a compact overall parameterization. We provide tractable inference and learning algorithms for arbitrary DST topologies via an efficient structured mean-field algorithm. The diverse applicability of DSTs is demonstrated by experiments on gene expression data and by modeling group behavior in the setting of an American football game.

cs.LG↗

arXiv2011

Kepler 18-b, c, and d: A System Of Three Planets Confirmed by Transit Timing Variations, Lightcurve Validation, Spitzer Photometry and Radial Velocity Measurements

We report the detection of three transiting planets around a Sunlike star, which we designate Kepler-18. The transit signals were detected in photometric data from the Kepler satellite, and were confirmed to arise from planets using a combination of large transit-timing variations, radial-velocity variations, Warm-Spitzer observations, and statistical analysis of false-positive probabilities. The Kepler-18 star has a mass of 0.97M_sun, radius 1.1R_sun, effective temperature 5345K, and iron abundance [Fe/H]= +0.19. The planets have orbital periods of approximately 3.5, 7.6 and 14.9 days. The innermost planet "b" is a "super-Earth" with mass 6.9 \pm 3.4M_earth, radius 2.00 \pm 0.10R_earth, and mean density 4.9 \pm 2.4 g cm^-3. The two outer planets "c" and "d" are both low-density Neptune-mass planets. Kepler-18c has a mass of 17.3 \pm 1.9M_earth, radius 5.49 \pm 0.26R_earth, and mean density 0.59 \pm 0.07 g cm^-3, while Kepler-18d has a mass of 16.4 \pm 1.4M_earth, radius 6.98 \pm 0.33R_earth, and mean density 0.27 \pm 0.03 g cm^-3. Kepler-18c and Kepler-18d have orbital periods near a 2:1 mean-motion resonance, leading to large and readily detected transit timing variations.

astro-ph.EP↗

arXiv2011

Planet Hunters: The First Two Planet Candidates Identified by the Public using the Kepler Public Archive Data

Planet Hunters is a new citizen science project, designed to engage the public in an exoplanet search using NASA Kepler public release data. In the first month after launch, users identified two new planet candidates which survived our checks for false- positives. The follow-up effort included analysis of Keck HIRES spectra of the host stars, analysis of pixel centroid offsets in the Kepler data and adaptive optics imaging at Keck using NIRC2. Spectral synthesis modeling coupled with stellar evolutionary models yields a stellar density distribution, which is used to model the transit orbit. The orbital periods of the planet candidates are 9.8844 \pm0.0087 days (KIC 10905746) and 49.7696 \pm0.00039 (KIC 6185331) days and the modeled planet radii are 2.65 and 8.05 R\oplus. The involvement of citizen scientists as part of Planet Hunters is therefore shown to be a valuable and reliable tool in exoplanet detection.

astro-ph.EP↗

arXiv2011

Spin-Orbit Alignment for the Circumbinary Planet Host Kepler-16A

Kepler-16 is an eccentric low-mass eclipsing binary with a circumbinary transiting planet. Here we investigate the angular momentum of the primary star, based on Kepler photometry and Keck spectroscopy. The primary star's rotation period is 35.1 +/- 1.0 days, and its projected obliquity with respect to the stellar binary orbit is 1.6 +/- 2.4 degrees. Therefore the three largest sources of angular momentum---the stellar orbit, the planetary orbit, and the primary's rotation---are all closely aligned. This finding supports a formation scenario involving accretion from a single disk. Alternatively, tides may have realigned the stars despite their relatively wide separation (0.2 AU), a hypothesis that is supported by the agreement between the measured rotation period and the "pseudosynchronous" period of tidal evolution theory. The rotation period, chromospheric activity level, and fractional light variations suggest a main-sequence age of 2-4 Gyr. Evolutionary models of low-mass stars can match the observed masses and radii of the primary and secondary stars to within about 3%.

astro-ph.EP↗

arXiv2011

M2K: II. A Triple-Planet System Orbiting HIP 57274

Doppler observations from Keck Observatory have revealed a triple planet system orbiting the nearby mid-type K dwarf, HIP 57274. The inner planet, HIP 57274b, is a super-Earth with \msini\ = 11.6 \mearth (0.036 \mjup), an orbital period of 8.135 $\pm$ 0.004 d, and slightly eccentric orbit $e=0.19 \pm 0.1$. We calculate a transit probability of 6.5% for the inner planet. The second planet has \msini\ = 0.4 \mjup\ with an orbital period of 32.0 $\pm 0.02$ d in a nearly circular orbit, and $e = 0.05 \pm 0.03$. The third planet has \msini\ = 0.53 \mjup\ with an orbital period of 432 $\pm 8$ d (1.18 years) and an eccentricity $e = 0.23 \pm 0.03$. This discovery adds to the number of super Earth mass planets with $\msini < 12 \mearth\ $ that have been detected with Doppler surveys. We find that 56 $\pm 18$% super-Earths are members of multi-planet systems. This is certainly a lower limit because of observational detectability limits, yet significantly higher than the fraction of Jupiter mass exoplanets, $20 \pm 8$%, that are members of Doppler-detected, multi-planet systems.

astro-ph.EP↗