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 & 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.
Asteroseismology probes the internal structures of stars by using their natural pulsation frequencies
. It relies on identifying sequences of pulsation modes that can be compared with theoretical ...models, which has been done successfully for many classes of pulsators, including low-mass solar-type stars
, red giants
, high-mass stars
and white dwarfs
. However, a large group of pulsating stars of intermediate mass-the so-called δ Scuti stars-have rich pulsation spectra for which systematic mode identification has not hitherto been possible
. This arises because only a seemingly random subset of possible modes are excited and because rapid rotation tends to spoil regular patterns
. Here we report the detection of remarkably regular sequences of high-frequency pulsation modes in 60 intermediate-mass main-sequence stars, which enables definitive mode identification. The space motions of some of these stars indicate that they are members of known associations of young stars, as confirmed by modelling of their pulsation spectra.
We carry out an independent search of Kepler photometry for small transiting planets with sizes 0.5-8.0 times that Earth and orbital periods between 5 and 50 days, with the goal of measuring the ...fraction of stars harboring such planets. We use a new transit search algorithm, TERRA, optimized to detect small planets around photometrically quiet stars. We restrict our stellar sample to include the 12,000 stars having the lowest photometric noise in the Kepler survey, thereby maximizing the detectability of Earth-size planets. We report 129 planet candidates having radii less than 6 RsubE found in three years of Kepler photometry. We compute the occurrence of planets as a function of planet radius and period, correcting for the detection completeness as well as the geometric probability of transit, R/a. The occurrence of planets is consistent with constant from 2 RsubE toward 1 RsubE. This unexpected plateau in planet occurrence at 2 RsubE suggests distinct planet formation processes for planets above and below 2 RsubE.
We present the first results from our Keck program investigating the orbital architectures of planet-hosting multiple star systems. We combine our Keck/NIRC2 adaptive optics astrometry with ...multi-epoch Keck/HIRES RVs of all three stars to determine a precise orbit for the BC pair around A, given their empirically constrained masses. We measure minimal astrometric motion, but our RVs reveal significant orbital velocity and acceleration. We validate that the system is dynamically stable in its present configuration via n-body simulations. We find that the A-BC orbit and planetary orbits are likely aligned (98%) given that they both have edge-on orbits and misalignment induces precession of the planets out of transit. We conclude that the stars were likely on their current orbits during the epoch of planet formation, truncating the protoplanetary disk at ~2 AU.
Radial velocity (RV) detection of planets is hampered by astrophysical processes on the surfaces of stars that induce a stochastic signal, or "jitter," which can drown out or even mimic planetary ...signals. Here, we empirically and carefully measure the RV jitter of more than 600 stars from the California Planet Search sample on a star by star basis. As part of this process, we explore the activity-RV correlation of stellar cycles and include appendices listing every ostensibly companion-induced signal we removed and every activity cycle we noted. We then use precise stellar properties from Brewer et al. to separate the sample into bins of stellar mass and examine trends with activity and with evolutionary state. We find that RV jitter tracks stellar evolution and that in general, stars evolve through different stages of RV jitter: the jitter in younger stars is driven by magnetic activity, while the jitter in older stars is convectively driven and dominated by granulation and oscillations. We identify the "jitter minimum"-where activity-driven and convectively driven jitter have similar amplitudes-for stars between 0.7 and 1.7 M and find that more-massive stars reach this jitter minimum later in their lifetime, in the subgiant or even giant phases. Finally, we comment on how these results can inform future RV efforts, from prioritization of follow-up targets from transit surveys like TESS to target selection of future RV surveys.
We present precise Doppler measurements of four stars obtained during the past decade at Keck Observatory by the California Planet Survey (CPS). These stars, namely, HD 34445, HD 126614, HD 13931, ...and Gl 179, all show evidence for a single planet in Keplerian motion. We also present Doppler measurements from the Hobby-Eberly Telescope (HET) for two of the stars, HD 34445 and Gl 179, that confirm the Keck detections and significantly refine the orbital parameters. These planets add to the statistical properties of giant planets orbiting near or beyond the ice line, and merit follow-up by astrometry, imaging, and space-borne spectroscopy. Their orbital parameters span wide ranges of planetary minimum mass (M sin i = 0.38-1.9 M{sub Jup}), orbital period (P = 2.87-11.5 yr), semimajor axis (a = 2.1-5.2 AU), and eccentricity (e = 0.02-0.41). HD 34445 b (P = 2.87 yr, M sin i = 0.79 M{sub Jup}, e = 0.27) is a massive planet orbiting an old, G-type star. We announce a planet, HD 126614 Ab, and an M dwarf, HD 126614 B, orbiting the metal-rich star HD 126614 (which we now refer to as HD 126614 A). The planet, HD 126614 Ab, has minimum mass M sin i = 0.38 M{sub Jup} and orbits the stellar primary with period P = 3.41 yr and orbital separation a = 2.3 AU. The faint M dwarf companion, HD 126614 B, is separated from the stellar primary by 489 mas (33 AU) and was discovered with direct observations using adaptive optics and the PHARO camera at Palomar Observatory. The stellar primary in this new system, HD 126614 A, has the highest measured metallicity (Fe/H = +0.56) of any known planet-bearing star. HD 13931 b (P = 11.5 yr, M sin i = 1.88 M{sub Jup}, e = 0.02) is a Jupiter analog orbiting a near solar twin. Gl 179 b (P = 6.3 yr, M sin i = 0.82 M{sub Jup}, e = 0.21) is a massive planet orbiting a faint M dwarf. The high metallicity of Gl 179 is consistent with the planet-metallicity correlation among M dwarfs, as documented recently by Johnson and Apps.
We present evidence for a correlation between the observed properties of hot Jupiter emission spectra and the activity levels of the host stars measured using Ca II H and K emission lines. We find ...that planets with dayside emission spectra that are well-described by standard one-dimensional atmosphere models with water in absorption (HD 189733, TrES-1, TrES-3, WASP-4) orbit chromospherically active stars, while planets with emission spectra that are consistent with the presence of a strong high-altitude temperature inversion and water in emission orbit quieter stars. We estimate that active G and K stars have Lyman {alpha} fluxes that are typically a factor of 4-7 times higher than quiet stars with analogous spectral types and propose that the increased UV flux received by planets orbiting active stars destroys the compounds responsible for the formation of the observed temperature inversions. In this paper, we also derive a model-independent method for differentiating between these two atmosphere types using the secondary eclipse depths measured in the 3.6 and 4.5 {mu}m bands on the Spitzer Space Telescope and argue that the observed correlation is independent of the inverted/non-inverted paradigm for classifying hot Jupiter atmospheres.
Abstract
In 2017, the California-Kepler Survey (CKS) published its first data release (DR1) of high-resolution optical spectra of 1305 planet hosts. Refined CKS planet radii revealed that small ...planets are bifurcated into two distinct populations, super-Earths (smaller than 1.5
R
⊕
) and sub-Neptunes (between 2.0 and 4.0
R
⊕
), with few planets in between (the “radius gap”). Several theoretical models of the radius gap predict variation with stellar mass, but testing these predictions is challenging with CKS DR1 due to its limited
M
⋆
range of 0.8–1.4
M
⊙
. Here we present CKS DR2 with 411 additional spectra and derived properties focusing on stars of 0.5–0.8
M
⊙
. We found that the radius gap follows
R
p
∝
P
m
with
m
= −0.10 ± 0.03, consistent with predictions of X-ray and ultraviolet- and core-powered mass-loss mechanisms. We found no evidence that
m
varies with
M
⋆
. We observed a correlation between the average sub-Neptune size and
M
⋆
. Over 0.5–1.4
M
⊙
, the average sub-Neptune grows from 2.1 to 2.6
R
⊕
, following
R
p
∝
M
⋆
α
with
α
= 0.25 ± 0.03. In contrast, there is no detectable change for super-Earths. These
M
⋆
–
R
p
trends suggest that protoplanetary disks can efficiently produce cores up to a threshold mass of
M
c
, which grows linearly with stellar mass according to
M
c
≈ 10
M
⊕
(
M
⋆
/
M
⊙
). There is no significant correlation between sub-Neptune size and stellar metallicity (over −0.5 to +0.5 dex), suggesting a weak relationship between planet envelope opacity and stellar metallicity. Finally, there is no significant variation in sub-Neptune size with stellar age (over 1–10 Gyr), which suggests that the majority of envelope contraction concludes after ∼1 Gyr.
Abstract
We present a high-precision radial velocity (RV) survey of 719 FGKM stars, which host 164 known exoplanets and 14 newly discovered or revised exoplanets and substellar companions. This ...catalog updated the orbital parameters of known exoplanets and long-period candidates, some of which have decades-longer observational baselines than they did upon initial detection. The newly discovered exoplanets range from warm sub-Neptunes and super-Earths to cold gas giants. We present the catalog sample selection criteria, as well as over 100,000 RV measurements, which come from the Keck-HIRES, APF-Levy, and Lick-Hamilton spectrographs. We introduce the new RV search pipeline
RVSearch
(
https://california-planet-search.github.io/rvsearch/
) that we used to generate our planet catalog, and we make it available to the public as an open-source Python package. This paper is the first study in a planned series that will measure exoplanet occurrence rates and compare exoplanet populations, including studies of giant planet occurrence beyond the water ice line, and eccentricity distributions to explore giant planet formation pathways. We have made public all radial velocities and associated data that we use in this catalog.
Abstract
We used high-precision radial velocity measurements of FGKM stars to determine the occurrence of giant planets as a function of orbital separation spanning 0.03–30 au. Giant planets are more ...prevalent at orbital distances of 1–10 au compared to orbits interior or exterior of this range. The increase in planet occurrence at ∼1 au by a factor of ∼4 is highly statistically significant. A fall-off in giant planet occurrence at larger orbital distances is favored over models with flat or increasing occurrence. We measure
14.1
−
1.8
+
2.0
giant planets per 100 stars with semimajor axes of 2–8 au and
8.9
−
2.4
+
3.0
giant planets per 100 stars in the range 8–32 au, a decrease in occurrence with increasing orbital separation that is significant at the ∼2
σ
level. We find that the occurrence rate of sub-Jovian planets (0.1–1 Jupiter masses) is also enhanced for 1–10 au orbits. This suggests that lower-mass planets may share the formation or migration mechanisms that drive the increased prevalence near the water–ice line for their Jovian counterparts. Our measurements of cold gas giant occurrence are consistent with the latest results from direct imaging surveys and gravitational lensing surveys despite different stellar samples. We corroborate previous findings that giant planet occurrence increases with stellar mass and metallicity.