Deep learning techniques have been well explored in the transiting exoplanet field, however previous work mainly focuses on classification and inspection. In this work, we develop a novel detection algorithm based on a well-proven object detection fr
amework in the computer vision field. Through training the network on the light curves of the confirmed Kepler exoplanets, our model yields 94% precision and 95% recall for transits with signal-to-noise ratio higher than 6 (set the confidence threshold to 0.6). Giving a slightly lower confidence threshold, recall can reach higher than 97%, which makes our model applicable for large-scale search. We also transfer the trained model to the TESS data and obtain similar performance. The results of our algorithm match the intuition of the human visual perception and make it easy to find single transiting candidates. Moreover, the parameters of the output bounding boxes can also help to find multiplanet systems. Our network and detection functions are implemented in the Deep-Transit toolkit, which is an open-source Python package hosted on GitHub and PyPI.
To fully constrain the orbits of low mass circumstellar companions, we conduct combined analyses of the radial velocity data as well as the Gaia and Hipparcos astrometric data for eight nearby systems. Our study shows that companion-induced position
and proper motion differences between Gaia and Hipparcos are significant enough to constrain orbits of low mass companions to a precision comparable with previous combined analyses of direct imaging and radial velocity data. We find that our method is robust to whether we use Gaia DR2 or Gaia EDR3, as well as whether we use all of the data, or just proper motion differences. In particular, we fully characterize the orbits of HD 190360 b and HD 16160 C for the first time. With a mass of 1.8$pm$0.2$m_{rm Jup}$ and an effective temperature of 123-176 K and orbiting around a Sun-like star, HD 190360 b is the smallest Jupiter-like planet with well-constrained mass and orbit, belonging to a small sample of fully characterized Jupiter analogs. It is separated from its primary star by 0.25$$ and thus may be suitable for direct imaging by the CGI instrument of the Roman Space Telescope.
We present a detailed study of the bulk electronic structure of high quality VSe$_{2}$ single crystals using optical spectroscopy. Upon entering the charge density wave phase below the critical temperature of 112 K, the optical conductivity of VSe$_2
$ undergoes a significant rearrangement. A Drude response present above the critical temperature is suppressed while a new interband transition appears around 0.07,eV. From our analysis, we estimate that part of the spectral weight of the Drude response is transferred to a collective mode of the CDW phase. The remaining normal state charge dynamics appears to become strongly damped by interactions with the lattice as evidenced by a mass enhancement factor m$^{*}$/m$approx$3. In addition to the changes taking place in the electronic structure, we observe the emergence of infrared active phonons below the critical temperature associated with the 4a x 4a lattice reconstruction.
Analysis of new precision radial velocity (RV) measurements from the Lick Automated Planet Finder (APF) and Keck HIRES have yielded the discovery of three new exoplanet candidates orbiting two nearby K dwarfs not previously reported to have companion
s (HD 190007 & HD 216520). We also report new velocities from both the APF and the Planet Finder Spectrograph (PFS) for the previously reported planet host stars GJ 686 and HD 180617 and update the corresponding exoplanet orbital models. Of the newly discovered planets, HD 190007 b has a period of 11.72 days, an RV semi-amplitude of K = 5.64$pm$0.55 m s$^{-1}$, a minimum mass of 16.46$pm$1.66 $rm M_{oplus}$, and orbits the slightly metal-rich, active K4 dwarf star HD 190007 (d = 12.7 pc). HD 216520 b has an orbital period of 35.45 days, an RV semi-amplitude of K = 2.28$pm$0.20 m s$^{-1}$, and a minimum mass of 10.26$pm$0.99 $rm M_{oplus}$, while HD 216520 c has an orbital period of P = 154.43 days, an RV semi-amplitude of K = 1.29$pm0.22$ m s$^{-1}$, and a minimum mass of 9.44$pm$1.63 $rm M_{oplus}$. Both of these planets orbit the slightly metal-poor, inactive K0 dwarf star HD 216520 (d = 19.6 pc). We find that our updated best fit models for HD 180617 b and GJ 686 b are in good agreement with the previously published results. For HD 180617 b we obtain an orbital period of 105.91 days, an RV semi-amplitude of K = 2.696$pm$0.22 m s$^{-1}$, and a minimum mass of 2.214$pm$1.05 $rm M_{oplus}$. For GJ 686 b we find the orbital period to be 15.53 days, the RV semi-amplitude to be K = 3.00$pm$0.18 m s$^{-1}$, and the minimum mass to be 6.624$pm$0.432 $rm M_{oplus}$. Using an injection-recovery exercise, we find that HD 190007 b and HD 216520 b are unlikely to have additional planets with masses and orbital periods within a factor of two, in marked contrast to $sim$85% of planets in this mass and period range found with Kepler.
Billions of text analysis requests containing private emails, personal text messages, and sensitive online reviews, are processed by recurrent neural networks (RNNs) deployed on public clouds every day. Although prior secure networks combine homomorp
hic encryption (HE) and garbled circuit (GC) to preserve users privacy, naively adopting the HE and GC hybrid technique to implement RNNs suffers from long inference latency due to slow activation functions. In this paper, we present a HE and GC hybrid gated recurrent unit (GRU) network, CryptoGRU, for low-latency secure inferences. CryptoGRU replaces computationally expensive GC-based $tanh$ with fast GC-based $ReLU$, and then quantizes $sigmoid$ and $ReLU$ with a smaller bit length to accelerate activations in a GRU. We evaluate CryptoGRU with multiple GRU models trained on 4 public datasets. Experimental results show CryptoGRU achieves top-notch accuracy and improves the secure inference latency by up to $138times$ over one of state-of-the-art secure networks on the Penn Treebank dataset.
We propose a neural network-based solution for three different tracks of 2nd International Illumination Estimation Challenge (chromaticity.iitp.ru). Our method is built on pre-trained Squeeze-Net backbone, differential 2D chroma histogram layer and a
shallow MLP utilizing Exif information. By combining semantic feature, color feature and Exif metadata, the resulting method -- SDE-AWB -- obtains 1st place in both indoor and two-illuminant tracks and 2nd place in general track.
Earth-sized planets in the habitable zones of M dwarfs are good candidates for the study of habitability and detection of biosignatures. To search for these planets, we analyze all available radial velocity data and apply four signal detection criter
ia to select the optimal candidates. We find ten strong candidates satisfying these criteria and three weak candidates showing inconsistency over time due to data samplings. We also confirm three previous planet candidates and improve their orbital solutions through combined analyses of updated data sets. Among the strong planet candidates, HIP 38594 b is a temperate super-Earth with a mass of $8.2 pm 1.7$ $M_oplus$ and an orbital period of $60.7pm0.1$ days, orbiting around an early-type M dwarf. Early-type M dwarfs are less active and thus are better hosts for habitable planets than mid-type and late-type M dwarfs. Moreover, we report the detection of five two-planet systems, including two systems made up of a warm or cold Neptune and a cold Jupiter, consistent with a positive correlation between these two types of planets. We also detect three temperate Neptunes, four cold Neptunes, and four cold Jupiters, contributing to a rarely explored planet population. Due to their proximity to the Sun, these planets on wide orbits are appropriate targets for direct imaging by future facilities such as HabEx and ELT.
We report the detection of a transiting, dense Neptune planet candidate orbiting the bright ($V=8.6$) K0.5V star HD 95338. Detection of the 55-day periodic signal comes from the analysis of precision radial velocities from the Planet Finder Spectrogr
aph on the Magellan II Telescope. Follow-up observations with HARPS also confirm the presence of the periodic signal in the combined data. HD 95338 was also observed by the Transiting Exoplanet Survey Satellite ({it TESS}) where we identify a clear single transit in the photometry. A Markov Chain Monte Carlo period search on the velocities allows strong constraints on the expected transit time, matching well the epoch calculated from tess{} data, confirming both signals describe the same companion. A joint fit model yields an absolute mass of 42.44$^{+2.22}_{-2.08} M_{oplus}$ and a radius of 3.89$^{+0.19}_{-0.20}$ $R_{oplus}$ which translates to a density of 3.98$^{+0.62}_{-0.64}$ gcm, for the planet. Given the planet mass and radius, structure models suggest it is composed of a mixture of ammonia, water, and methane. HD 95338,b is one of the most dense Neptune planets yet detected, indicating a heavy element enrichment of $sim$90% ($sim38, M_{oplus}$). This system presents a unique opportunity for future follow-up observations that can further constrain structure models of cool gas giant planets.
The radial velocity method plays a major role in the discovery of nearby exoplanets. To efficiently find planet candidates from the data obtained in high precision radial velocity surveys, we apply a signal diagnostic framework to detect radial veloc
ity signals that are statistically significant, consistent in time, robust to the choice of noise models, and not correlated with stellar activity. Based on the application of this approach to the survey data of the Planet Finder Spectrograph (PFS), we report fifteen planet candidates located in fourteen stellar systems. We find that the orbits of the planet candidates around HD 210193, 103949, 8326, and 71135 are consistent with temperate zones around these stars (where liquid water could exist on the surface). With periods of 7.76 and 15.14 days respectively, the planet candidates around star HIP 54373 form a 1:2 resonance system. These discoveries demonstate the feasibility of automated detection of exoplanets from large radial velocity surveys, which may provide a complete sample of nearby Earth analogs.
Yelp has been one of the most popular local service search engine in US since 2004. It is powered by crowd-sourced text reviews and photo reviews. Restaurant customers and business owners upload photo images to Yelp, including reviewing or advertisin
g either food, drinks, or inside and outside decorations. It is obviously not so effective that labels for food photos rely on human editors, which is an issue should be addressed by innovative machine learning approaches. In this paper, we present a simple but effective approach which can identify up to ten kinds of food via raw photos from the challenge dataset. We use 1) image pre-processing techniques, including filtering and image augmentation, 2) feature extraction via convolutional neural networks (CNN), and 3) three ways of classification algorithms. Then, we illustrate the classification accuracy by tuning parameters for augmentations, CNN, and classification. Our experimental results show this simple but effective approach to identify up to 10 food types from images.
