Search for: All records

We use Bayesian convolutional neural networks and a novel generative model of Galaxy Zoo volunteer responses to infer posteriors for the visual morphology of galaxies. Bayesian CNN can learn from galaxy images with uncertain labels and then, for previously unlabelled galaxies, predict the probability of each possible label. Our posteriors are well-calibrated (e.g. for predicting bars, we achieve coverage errors of 11.8 per cent within a vote fraction deviation of 0.2) and hence are reliable for practical use. Further, using our posteriors, we apply the active learning strategy BALD to request volunteer responses for the subset of galaxies which, if labelled, would be most informative for training our network. We show that training our Bayesian CNNs using active learning requires up to 35–60 per cent fewer labelled galaxies, depending on the morphological feature being classified. By combining human and machine intelligence, Galaxy zoo will be able to classify surveys of any conceivable scale on a time-scale of weeks, providing massive and detailed morphology catalogues to support research into galaxy evolution.

 

Many scientific domains gather sufficient labels to train machine algorithms through human-in-the-loop techniques provided by the this http URL citizen science platform. As the range of projects, task types and data rates increase, acceleration of model training is of paramount concern to focus volunteer effort where most needed. The application of Transfer Learning (TL) between Zooniverse projects holds promise as a solution. However, understanding the effectiveness of TL approaches that pretrain on large-scale generic image sets vs. images with similar characteristics possibly from similar tasks is an open challenge. We apply a generative segmentation model on two Zooniverse project-based data sets: (1) to identify fat droplets in liver cells (FatChecker; FC) and (2) the identification of kelp beds in satellite images (Floating Forests; FF) through transfer learning from the first project. We compare and contrast its performance with a TL model based on the COCO image set, and subsequently with baseline counterparts. We find that both the FC and COCO TL models perform better than the baseline cases when using >75% of the original training sample size. The COCO-based TL model generally performs better than the FC-based one, likely due to its generalized features. Our investigations provide important insights into usage of TL approaches on multi-domain data hosted across different Zooniverse projects, enabling future projects to accelerate task completion. 

ABSTRACT We present Galaxy Zoo DECaLS: detailed visual morphological classifications for Dark Energy Camera Legacy Survey images of galaxies within the SDSS DR8 footprint. Deeper DECaLS images (r = 23.6 versus r = 22.2 from SDSS) reveal spiral arms, weak bars, and tidal features not previously visible in SDSS imaging. To best exploit the greater depth of DECaLS images, volunteers select from a new set of answers designed to improve our sensitivity to mergers and bars. Galaxy Zoo volunteers provide 7.5 million individual classifications over 314 000 galaxies. 140 000 galaxies receive at least 30 classifications, sufficient to accurately measure detailed morphology like bars, and the remainder receive approximately 5. All classifications are used to train an ensemble of Bayesian convolutional neural networks (a state-of-the-art deep learning method) to predict posteriors for the detailed morphology of all 314 000 galaxies. We use active learning to focus our volunteer effort on the galaxies which, if labelled, would be most informative for training our ensemble. When measured against confident volunteer classifications, the trained networks are approximately 99 per cent accurate on every question. Morphology is a fundamental feature of every galaxy; our human and machine classifications are an accurate and detailed resource for understanding how galaxies evolve. 

Camera traps - remote cameras that capture images of passing wildlife - have become a ubiquitous tool in ecology and conservation. Systematic camera trap surveys generate ‘Big Data’ across broad spatial and temporal scales, providing valuable information on environmental and anthropogenic factors affecting vulnerable wildlife populations. However, the sheer number of images amassed can quickly outpace researchers’ ability to manually extract data from these images (e.g., species identities, counts, and behaviors) in timeframes useful for making scientifically-guided conservation and management decisions. Here, we present ‘Snapshot Safari’ as a case study for merging citizen science and machine learning to rapidly generate highly accurate ecological Big Data from camera trap surveys. Snapshot Safari is a collaborative cross-continental research and conservation effort with 1500+ cameras deployed at over 40 eastern and southern Africa protected areas, generating millions of images per year. As one of the first and largest-scale camera trapping initiatives, Snapshot Safari spearheaded innovative developments in citizen science and machine learning. We highlight the advances made and discuss the issues that arose using each of these methods to annotate camera trap data. We end by describing how we combined human and machine classification methods (‘Crowd AI’) to create an efficient integrated data pipeline. Ultimately, by using a feedback loop in which humans validate machine learning predictions and machine learning algorithms are iteratively retrained on new human classifications, we can capitalize on the strengths of both methods of classification while mitigating the weaknesses. Using Crowd AI to quickly and accurately ‘unlock’ ecological Big Data for use in science and conservation is revolutionizing the way we take on critical environmental issues in the Anthropocene era. 

ABSTRACT We report on the discovery and validation of a two-planet system around a bright (V  = 8.85 mag) early G dwarf (1.43  R⊙, 1.15  M⊙, TOI 2319) using data from NASA’s Transiting Exoplanet Survey Satellite (TESS). Three transit events from two planets were detected by citizen scientists in the month-long TESS light curve (sector 25), as part of the Planet Hunters TESS project. Modelling of the transits yields an orbital period of $11.6264 _{ - 0.0025 } ^ { + 0.0022 }$ d and radius of $3.41 _{ - 0.12 } ^ { + 0.14 }$ R⊕ for the inner planet, and a period in the range 19.26–35 d and a radius of $5.83 _{ - 0.14 } ^ { + 0.14 }$ R⊕ for the outer planet, which was only seen to transit once. Each signal was independently statistically validated, taking into consideration the TESS light curve as well as the ground-based spectroscopic follow-up observations. Radial velocities from HARPS-N and EXPRES yield a tentative detection of planet b, whose mass we estimate to be $11.56 _{ - 6.14 } ^ { + 6.58 }$ M⊕, and allow us to place an upper limit of 27.5 M⊕ (99 per cent confidence) on the mass of planet c. Due to the brightness of the host star and the strong likelihood of an extended H/He atmosphere on both planets, this system offers excellent prospects for atmospheric characterization and comparative planetology. 

Abstract We present the results from the first two years of the Planet Hunters TESS citizen science project, which identifies planet candidates in the TESS data by engaging members of the general public. Over 22,000 citizen scientists from around the world visually inspected the first 26 Sectors of TESS data in order to help identify transit-like signals. We use a clustering algorithm to combine these classifications into a ranked list of events for each sector, the top 500 of which are then visually vetted by the science team. We assess the detection efficiency of this methodology by comparing our results to the list of TESS Objects of Interest (TOIs) and show that we recover 85 % of the TOIs with radii greater than 4 ⊕ and 51 % of those with radii between 3 and 4 R⊕. Additionally, we present our 90 most promising planet candidates that had not previously been identified by other teams, 73 of which exhibit only a single transit event in the TESS light curve, and outline our efforts to follow these candidates up using ground-based observatories. Finally, we present noteworthy stellar systems that were identified through the Planet Hunters TESS project. 

ABSTRACT We report on the discovery and validation of TOI 813 b (TIC 55525572 b), a transiting exoplanet identified by citizen scientists in data from NASA’s Transiting Exoplanet Survey Satellite (TESS) and the first planet discovered by the Planet Hunters TESS project. The host star is a bright (V = 10.3 mag) subgiant ($R_\star =1.94\, R_\odot$, $M_\star =1.32\, M_\odot$). It was observed almost continuously by TESS during its first year of operations, during which time four individual transit events were detected. The candidate passed all the standard light curve-based vetting checks, and ground-based follow-up spectroscopy and speckle imaging enabled us to place an upper limit of $2\, M_{\rm Jup}$ (99 per cent confidence) on the mass of the companion, and to statistically validate its planetary nature. Detailed modelling of the transits yields a period of $83.8911 _{ - 0.0031 } ^ { + 0.0027 }$ d, a planet radius of 6.71 ± 0.38 R⊕ and a semimajor axis of $0.423 _{ - 0.037 } ^ { + 0.031 }$ AU. The planet’s orbital period combined with the evolved nature of the host star places this object in a relatively underexplored region of parameter space. We estimate that TOI 813 b induces a reflex motion in its host star with a semi-amplitude of ∼6 m s−1, making this a promising system to measure the mass of a relatively long-period transiting planet.