Some notes about marrying astrophysics with modern machine learning with particular focus on deep learning.

DeepSpace.md

Machine learning and deep learning in astronomy:

Machine learning has a long history in astronomy, but deep learning (DL) only got traction since arount 2016. Here I will list current DL effords:

The main field of ML applications in astrophysics is object classification. With source counts now ranging into the 10⁷-10⁸ for most surveys, machine learning is put to use to allow the classification of a large number of sources which would otherwise need an infeasible amount of manpower:

• Classifying galaxies according to their HI content - SVM perform best in specisifity and is hence used over a DNN
• Morphological classification of radio galaxies: Capsule Networks versus Convolutional Neural Networks *05/2019 - CNN show better performance than Capsule Networks, dropout provides more robusteness against image noise. Trained with LoTTs images.
• Optical Transient Object Classification in Wide Field Small Aperture Telescopes with Neural Networks 05/2019 - CNN vs LSTMs, in this case (after Sextrator preselection) CNNs perform better
• Identifying MgII Narrow Absorption Lines with Deep Learning 04/2019 - simple 2DCNN (3 conv. blocks), training with real sata and simulations
• Identifying Galaxy Mergers in Observations and Simulations with Deep Learning 02/2019 - ROC curve shows that many merger cases are hard to trace
• Machine and Deep Learning Applied to Galaxy Morphology - A Complete Classification Catalog 01/2019 - tested both traditional ML and DL
• Classification of Broad Absorption Line Quasars with a Convolutional Neural Network 01/2019 - used very shallow CNN (1 conv. layer) with good results, because of simple classification problem on data with PCA-subtracted continuum
• A MACHINE LEARNING BASED MORPHOLOGICAL CLASSIFICATION OF 14,251 RADIO AGNS SELECTED FROM THE BEST-HECKMAN SAMPLE 12/2018 - smart pretraining through unsuperivesed autoencoder, fine-tuning with labeled sample, and classification through three on the (deep?) features of the CNN
• Multiband galaxy morphologies for CLASH: a convolutional neural network transferred from CANDELS 10/2018 - fine tuning a DCNN for a similar domain
• Galaxy morphology prediction using capsule networks 09/2018 - uses Capsule Networks that outperform the baseline model by a large margin. Nice!
• Galaxy detection and identification using deep learning and data augmentation 09/2018 - YOLO + OpenCV
• QuasarNET: Human-level spectral classification and redshifting with Deep Neural Networks 08/2018 - human level means expert level with a two-stage DL approach.
• Detecting Radio Frequency Interference in radio-antenna arrays with the Recurrent Neural Network algorithm 08/2018 High precision, low efficiency (many false positives) with LSTM
• Galaxy Morphology Classification with Deep Convolutional Neural Networks 07/2018 - Great explaination of the preopocessing pipeline)
• The FIRST Classifier: Compact and Extended Radio Galaxy Classification using Deep Convolutional Neural Networks 07/2018 - very slim (only three conv layers) and great results for 5 classes
• DeepSource: Point Source Detection using Deep Learning 07/2018
• Radio Galaxy Zoo: ClaRAN — a deep learning classifier for radio morphologies 2018
• Using transfer learning to detect galaxy mergers 2018 - They also employ isotonic regression to get a class score that can be interpreted as probabiltiy.
• Machine-learning identification of extragalactic objects in the optical-infrared all-sky surveys 2018
• Lunar Crater Identification via Deep Learning 2018
• Deep Learning Classification in Asteroseismology Using an Improved Neural Network: Results on 15000 Kepler Red Giants and Applications to K2 and TESS Data 2018
• Radio Galaxy Zoo: Compact and extended radio source classification with deep learning
• Matching matched filtering with deep networks for gravitational-wave astronomy 2018
• Improving galaxy morphologies for SDSS with Deep Learning 2017
• Classifying Complex Faraday Spectra with Convolutional Neural Networks 2017
• ASSESSING THE PERFORMANCE OF A MACHINE LEARNING ALGORITHM IN IDENTIFYING BUBBLES IN DUST EMISSION 2017
• Parameterizing Stellar Spectra Using Deep Neural Networks 2017
• Effective Image Differencing with ConvNets for Real-time Transient Hunting 2017
• An Application of Deep Neural Networks in the Analysis of Stellar Spectra 2017
• Classifying Radio Galaxies with Convolutional Neural Network 2017; githublink
• Deep Galaxy: Classification of Galaxies based on Deep Convolutional Neural Networks
• Galaxy Classifications with Deep Learning
• Star-galaxy Classification Using Deep Convolutional Neural Networks 2016
• The ASKAP/EMU Source Finding Data Challenge 2015

The shades range from rich data sets like these to identifying rare events or objects in highly noisy/growded data, often data streams. Gravitational waves, and gravitational lenses are some of the favorites:

• An extended catalog of galaxy-galaxy strong gravitational lenses discovered in DES using convolutional neural networks *05/2019 - combined two classifiers using 4 InceptionsLayers (4filters) and trained on simulations
• Predicting Solar Flares Using a Long Short-Term Memory Network 05/2019 - LSTM and RF perform best dpeending on the classes, as input using a feature summary on time-step basis.
• Using Convolutional Neural Networks to identify Gravitational Lenses in Astronomical images - shallow, four convolutional layers
• Application of Machine Learning to the Particle Identification of GAPS 05/2019 - baloon survey of cosmic-ray antiparticles; first experiences with simple deep FCNN
• Detecting Exoplanet Transits through Machine LearningTechniques with Convolutional Neural Networks 05/2019 - 2D-CNN seem to work well with the transit-phase folding technique
• IDENTIFYING EXOPLANETS WITH DEEP LEARNING III: AUTOMATED TRIAGE AND VETTING OFTESSCANDIDATES 04/2019 - triage and vetting mode both from joined CNN branches
• RAPID: Early Classification of Explosive Transients using Deep Learning 04/2019 - generall purpose software, leveraging Gating Recurrent Units (GRUs)
• IDENTIFYING EXOPLANETS WITH DEEP LEARNING II: TWO NEW SUPER-EARTHS UNCOVERED BY ANEURAL NETWORK IN K2 DATA 03/2019 - using a three-rooted CNN to incorporate several data sources. They create a probability score with great success
• Classifying the unknown: discovering novel gravitational-wave detector glitches using similarity learning 03/2019 - used to improve the citizen-science project Gravity Spy
• DASH: DEEP LEARNING FOR THE AUTOMATED SPECTRAL CLASSIFICATION OF SUPERNOVAE ANDTHEIR HOSTS 03/2019 - two layered 2DCNN + fully connected. Simple structure, good results because of preprocessing. What is odd that they reshape their 1024 spectrum to 32x32. Plus is that they also developed an GUI.
• Rapid Classification of TESS Planet Candidates with ConvolutionalNeural Networks 02/2019 - using 2 1D-CNN and adding additional information in the FC-part
• Deep learning detection of transients 02/2019 - LSTM detections of transients in image-series, good paper
• Machine Vision and Deep Learning for Classification of Radio SETI Signals 02/2019 - Tested XGboost, Decision Trees, ResNet and DenseNet ... generally good performancy for all cases except the low(est) signal-noise cases that they have algorithms for
• Deep Learning for Multi-Messenger AstrophysicsA Gateway for Discovery in the Big Data Era 02/2019 - Focused on the detection of gravitational waves, but also kind of a review/vision paper
• Classification and Recovery of Radio Signals from Cosmic Ray Induced Air Showers with Deep Learning - simple 1D-DCNN feed with data denoised by a deep autoencoder (with some skip connections), trained on simulated data
• LinKS: Discovering galaxy-scale strong lenses in the Kilo-Degree Survey using Convolutional Neural Networks 12/2018 - paper from series on LinKS, results of combining three networks
• Gamma/Hadron Separation in Imaging Air Cherenkov Telescopes Using Deep Learning Libraries TensorFlow and PyTorch 11/2018 - Study still limited, further improvements exspected through more elaborate CNN and consideration of hexagonal resolution element shape
• Identification of Low Surface Brightness Tidal Features in Galaxies Using Convolutional Neural Networks 11/2018 - two CNN are combined to a stronger predictor
• Machine-learning Approaches to Exoplanet Transit Detection and Candidate Validation in Wide-field Ground-based Surveys RFC and CNN combined
• Finding high-redshift strong lenses in DES using convolutional neural networks 11/2018
• Scientific Domain Knowledge Improves Exoplanet Transit Classification with Deep Learning *10/21018 - implementing prior assumptions and two-scale abbraoch (local, global) to make the Classificator smaller & better
• Deep multi-survey classification of variable stars
• [Improved Photometric Classification of Supernovae using Deep Learning] - improved classification through introduction of timed recurrent unit and better data augmentation
• Graph Neural Networks for IceCube Signal Classification
• Deep Learning Based Detection of Cosmological Diffuse Radio Sources 09/2018 - CNN on tiles of simulated images to train for detection of faint diffuse emission
• Towards online triggering for the radio detection of air showers using deep neural networks 09/2018
• Stellar Cluster Detection using GMM with Deep Variational Autoencoder 09/2018
• Searching for Short Duration Stellar Variability with Wide-Field Star Trails and Deep Learning 08/2018 - They use a Unet + CNN
• Enhanced Rotational Invariant Convolutional Neural Network for Supernovae Detection 08/2018
• Deep Learning for Image Sequence Classification of Astronomical Events
• SEARCHING FOR HOT SUBDWARF STARS FROM THE LAMOST SPECTRA. III. CLASSIFYING THE HOT SUBDWARF STARS FROM FOURTH DATA RELEASE OF LAMOST USING DEEP LEARNING METHOD. 2018
• DETECTING SOLAR-LIKE OSCILLATIONS IN RED GIANTS WITH DEEP LEARNING 2018
• Application of Deep Learning methods to analysis of Imaging tmospheric Cherenkov Telescopes data
• Image-based deep learning for classification of noise transients in gravitational wave detectors 2018
• Testing Convolutional Neural Networks for finding strong gravitational lenses in KiDS 06/2018
• Classification of simulated radio signals using Wide Residual Networks for use in the search for extra-terrestrial intelligence 2018
• Characterizing the velocity of a wandering black hole and properties of the surrounding medium using convolutional neural networks
• APPLYING DEEP LEARNING TO FAST RADIO BURST CLASSIFICATION 2018 - They use a hyprid DNN system of 3 convolutional nets in different parameter spaces and a fully connected feed-forward net
• Identifying Exoplanets with Deep Learning: A Five Planet Resonant Chain around Kepler-80 and an Eighth Planet around Kepler-90 2017
• Supervised detection of exoplanets in high-contrast imaging sequences 2017
• Single-epoch supernova classification with deep convolutional neural networks 2017
• A Method Of Detecting Gravitational Wave Based On Time-frequency Analysis And Convolutional Neural Networks 2017
• Glitch Classification and Clustering for LIGO with Deep Transfer Learning 2017

as we see there modern approaches combining heterogeneous kind of data like

• A hybrid approach to machine learning annotation of large galaxy image databases 10/2018
• Retrieving Exoplanetary Atmospheres 2018
• Photometric redshift estimation via deep learning 2017

In some cases they are used to inform the researcher about novelties/anomalies

• Neural network-based anomaly detection for high-resolution X-ray spectroscopy 06/2019 - Using a VAE on spectra (7800pixels) with FC-blocks. Applying PCA on the 8 components of the innermost layer
• Deep Neural Network Classifier for Variable Stars with Novelty Detection Capability 05/2019 - A joint training of a two-headed classification and autoencoder-network allows for mutual improvements, classification and novelty detection

about shortcomings of their own metrices/features like

• Do Androids Dream of Magnetic Fields? Using Neural Networks to Interpret the Turbulent Interstellar Medium 05/2019

Other project investigate more generall purpose deep learning applications like:

• Accelerated Bayesian inference using deep learning 04/2019 - can be used for better proposals of multi-modal distributions in multidimensional spaces
• HexagDLy - Processing hexagonally sampled data withCNNs in PyTorch 03/2019 - they created the software to allow for that
• Transfer Learning in Astronomy: A New Machine-Learning Paradigm 12/2018
• CosmoFlow: Using Deep Learning to Learn the Universe at Scale 08/2018 - which did tremendous work to put DL on scale for computing clusters
• SBAF: A New Activation Function for Artificial Neural Net based Habitability Classification *2018
• Automatic physical inference 2018
• Eventsheduling for SWIFT Transients 2017
• J. Akeret programmed a tensorflow U-net Radio frequency interference mitigation using deep convolutional neural networks 2016

and regression/modeling/inversion (i.e. parametric predictions) of point spread functions

• Resolution and accuracy of non-linear regression of PSF with artificial neural networks 2018 - 5 conv layers+ 2 FC layers to predict 72 zernice polynomials
• Fast Point Spread Function Modeling with Deep Learning 2018 and other models:
• Deconfusing intensity maps with neural networks 05/2019 - using a ResBlocks architecture and train on sopihisticated simulations. They find that the accuracy drops significantly for cases where the net was not trained on. They ask if the training by simulation appproach should be treated with caution due to this possible biases. Will cosmology ever harness the full potential of DL?
• Projected Pupil Plane Pattern (PPPP) with artificialNeural Networks 05/2019
• A deep learning model to emulate simulations of cosmicreionization 05/2019
• Fast Wiener filtering of CMB maps with Neural Networks 05/2019 - Cool stuff, generating suited training data not through Wiener filtered images, but nifty Loss functions that ensure the Wiener-filter criterion. Using UNet, 1000 times speedup compared to exact Wiener filter
• Stokes Inversion based on Convolutional Neural Networks 04/2019 - succesion paper of this providing more insight plus another 'concatanete'-NN structure
• Baryon density extraction and isotropy analysis of Cosmic Microwave Background using a Multilayer perceptron 04/2019 - simple MLP with 5 hidden layers, results not overwhelming
• Painting with baryons: augmentingN-body simulationswith gas using deep generative models 03/2019
• Automatic detection of Interplanetary Coronal Mass Ejections from in-situ data: a deep learning approach 03/2019 - they developed a pipeline (including a CNN) that provides an automatic ICME detection from the WIND spacecraft in-situ measurements
• Learning the Relationship between Galaxies Spectra andtheir Star Formation Histories using Convolutional NeuralNetworks and Cosmological Simulations 03/2019 - good results on spectra with simple two-layered CNNs
• Mapping neutron star data to the equation of state of the densestmatter using the deep neural network 03/2019 - Using only a 5-layer NN
• Deep Learning at Scale for Gravitational Wave Parameter Estimationof Binary Black Hole Mergers 03/2019 - they use a socalled Root-Network architecture
• Simultaneous calibration of spectro-photometric distances and the GaiaDR2 parallax zero-point offset with deep learning 02/2019
• Separating the EoR Signal with a Convolutional DenoisingAutoencoder: A Deep-learning-based Method 02/2019 - applying a 8-layer CNN on a FTT sky distribution they are able beat all other tequniques by a huge margin.
• Galaxy shape measurement with convolutional neuralnetworks 02/2019 - using a custom CNN architecture to measure ellipticity with 4 orders of magnitude higher speed and higher precision.
• Convolutional Neural Networks on the HEALPix sphere: apixel-based algorithm and its application to CMB data analysis 02/2019
• [RADYNVERSION: Learning to Invert a Solar Flare Atmosphere with Invertible Neural Networks](https://arxiv.org/pdf/1901.08626.pdf#cite.Ardizzone2018 *01/2018) - using this paper to infer spectral line data
• Photometric Redshift Analysis using Supervised Learning Algorithms and Deep Learning 01/2019
• Gravitational Wave Denoising of Binary Black Hole Mergers with Deep Learning 01/2019
• Denoising Weak Lensing Mass Maps with Deep Learning 12/2018 - Using conditional GAN, i.e. Googles pix2pix
• Real-time multiframe blind deconvolution of solar images *12/2018 - Unet like Encode-Decoder structure with additional blocks and skip connections in them instead of layers, great results
• Particle Identification in Ground-Based Gamma-Ray Astronomy Using Convolutional Neural Networks 12/2018 - No real description of the used CNNs provided
• Deep Learning Applied to the Asteroseismic Modeling of Stars with Coherent Oscillation Modes 11/2018 - Deep Learming prediction as starting point for more precise parameter search
• Applying deep neural networks to the detection and space parameter estimation of compact binary coalescence with a network of gravitational wave detectors 11/2018
• On the dissection of degenerate cosmologies with machine learning 10/2018
• Distinguishing standard and modified gravity cosmologies with machine learning 10/2018
• Deblending galaxy superpositions with branched generative adversarial networks 10/2018
• DeepCMB: Lensing Reconstruction of the Cosmic Microwave Background with Deep Neural Networks - using ResUNet & SELU)
• Weak lensing shear estimation beyond the shape-noise limit: a machine learning approach 08/2018
• Analyzing Interferometric Observations of Strong Gravitational Lenses with Recurrent and Convolutional Neural Networks 08/2018
• Cosmological constraints from noisy convergence maps through deep learning
• Identifying Reionization Sources from 21cm Maps using Convolutional Neural Networks
• Weak-lensing shear measurement with machine learning
• Photometric redshifts from SDSS images using a Convolutional Neural Network 2018
• Deep learning from 21-cm images of the Cosmic Dawn 2018
• MADE: Improved Mass, Age, and Distance Estimates with Bayesian machine learning 2018
• Predicting the Neutral Hydrogen Content of Galaxies From Optical Data Using Machine Learning 2018
• Reionization Models Classifier using 21cm Map Deep Learning 2017
• Towards understanding feedback from supermassive black holes using convolutional neural networks 2017
• Redshift regression (add paper k. polsterer et al.) sometimes even with error estimates:
• An Ensemble of Bayesian Neural Networks for Exoplanetary Atmospheric Retrieval 05/2019 - GREAT STUFF1 I learned a lot! They use an ensemble of five Bayesian networks, each consisting of 4 Dense Concrete Dropout Layers (Making the Dropout learnable too), use the negative log-likelyhood as loss and get atmospheric retrieval with an error estimate.

as well as speed-up of simulations (i.e. a generative approach):

• Learning Radiative Transfer Models for Climate Change Applications inImaging Spectroscopy 06/2019
• HIGAN: COSMIC NEUTRAL HYDROGEN WITH GENERATIVE ADVERSARIAL NETWORKS 05/2019 - 'WGAN can produce new samples five orders of magnitude faster than hydrodynamic simulations'
• A volumetric deep Convolutional Neural Network for simulation of dark matter halo catalogues 2018
• Fast Cosmic Web Simulations with Generative Adversarial Networks
• Painting galaxies into dark matter halos using machine learning 2017

synthetic data generation:

• A Halo Merger Tree Generation and Evaluation Framework 06/2019 - GAN of MergerTree properties (Mass,distiance,etc.) in matrix representation with column and row wise filters with architecture based on DCGAN
• RadioGAN – Translations between different radio surveyswith generative adversarial networks 06/2019
• Forging new worlds: high-resolution synthetic galaxies with chained generative adversarial networks 11/2018
• Generative deep fields: arbitrarily sized, random synthetic astronomical images through deep learning 04/2019 - simpe GAN, but using spectral normalizazion, ELU

retraining networks for classification:

• Transfer learning for radio galaxy classification 03/2018 - tested for surveys NVSS and FIRST, and more is planned
• Knowledge transfer of Deep Learning for galaxy morphology from one survey to another 2017

denoising:

• Advanced signal reconstruction in Tunka-Rex 06/2019 - using a deep 1D-convolutional-AE
• DENOISING GRAVITATIONAL WAVES WITH ENHANCED DEEP RECURRENT DENOISING AUTO-ENCODERS 03/2019 - By introducing additional structuresto the model (cross-layer connection, signal amplifier), and by curriculum learning, their model outperforms tested popular denoising algorithms (PCA, dictionary learning and wavelet thresholding) for GW denoising.
• Real-time multiframe blind deconvolution of solar images 2018 - with super resolution in video frames

catalogue creation by unsupervised clustering:

• Radio Galaxy Zoo: Unsupervised Clustering of Convolutionally Auto-encoded Radio-astronomical Images *06/2019 - SOM on PCA + Deep-Conv-Autoencode features of pre-augmented images (to cluster more due to source classes and not image features)
• Unsupervised learning and data clustering for the construction of Galaxy Catalogs in the Dark Energy Surveys 12/2018 - tSNe + Xception recursive training

semantic segmentation:

• Morpheus: A Deep Learning Framework For Pixel-Level Analysis of Astronomical Image Data 06/2019 - using a U-Net structure consisting of simple blocks and with a combined classification by the sliding window abbroach.
• Automated crater shape retrieval using weakly-supervised deep learning 06/2019 - instance segmentation using MaskRCNN
• Optimizing Sparse RFI Prediction using Deep Learning *02/2019 - good results, also considering the complex phase in a parallel network with concatantion at several joints. General a U-depth structure but with networks in networks. Note that they might become even better by introducing complex convolutions like in MRI
• Segmentation of coronal holes in solar disk images with a convolutional neural network 9/2018 Metastudies:
• Galaxy classification:A machine learning analysis of GAMA catalogue data 03/2019 - [sic] based on Generalized Relevance MatrixLearning Vector Quantization and Random Forests – we find that neither thedata from the individual catalogues nor a combined dataset based on all 5 cat-alogues fully supports the visual-inspection-based galaxy classification schemeemployed to categorise the galaxies.
• The Hubble Sequence at z∼0 in the IllustrisTNGsimulation with deep learning 03/2019
• Visualizing the Hidden Features of Galaxy Morphology with Machine Learning 2018 testing three different CN architectures and tSNe investigation of the deeo features including some reclassification

active learning:

• https://arxiv.org/pdf/1905.08200.pdf 05/2019 - Leveraging a CNN-Bayesian network as a proposal network for (re)classification in citizen-science thereby decreasing the needed number of samples by a large margin.

or even mission planning:

• Trajectory Design of Multiple Near Earth Asteroids Exploration Using Solar Sail Based on Deep Neural Network - TOP NOTCH. A combination of deep dense NN + Monte Carlo Tree Search let them solve the target selection and optimal sequence for the mult-asteroid exploration problem Some projects explicity aim to help the human in the loop, by providing preordered and ranked data for classification:
• Modeling with the Crowd:Optimizing the Human-Machine Partnershipwith Zooniverse 03/2019

In addition to that, there is a large fundus of more traditional, already established ML methods for classification (random tree, XGboost, SVM, filters like Kalman, Bayesian shallow NN), dimensionality reduction (t-sne or Cohen maps and neural gas), regression and in fact most of the applications mentioned above, like:

• A Principal Component Analysis-based method to analyse high-resolution spectroscopic data 06/2019 - applying a cross-correlation function on iamges with the first (+second?) PCA subtracted (being better than median subtraction).
• A machine learning approach for GRB detection in AstroSat CZTIdata - dynamic time wrapping (DTW, aleviating the need for feature extraction or a dense template) with symmetric Kullbback-Leibler metric & DBSCAN; nifty!
• AUTOMATIC CLASSIFICATION OF K2 PULSATING STARS USING MACHINELEARNING TECHNIQUES 06/2019 - posed as a classification problem using photometric multiband features; using a RF.
• DETERMINING SURFACE ROTATION PERIODS OF SOLAR-LIKE STARSOBSERVED BY THE KEPLER MISSION USING MACHINE LEARNING TECHNIQUES 06/2019 - posed as a classification problem on Autocorrelation and Spectrogramm features; using a RF.
• Constraining strongly-coupled new physics from cosmic rays withmachine learning techniques 06/2019 - using linear discriminant analysis (LDA), a quadratic discriminant anal-ysis (QDA), a SVM and a MLP on binned data
• An interpretable machine learning framework for darkmatter halo formation 06/2019 - using GradientBoosting
• General classification of light curves using extremeboosting 06/2019 - using XGBoost
• One simulation to have them all: performance of the Bias Assignment Method against N-body simulations 06/2019 - not deep learning (yet), but still a scifi optimitation method to my mind
• [k-Means Aperture Optimization Applied to Kepler K2 Time Series Photometry of Titan 06/2019]
• KiDS-SQuaD II: Machine learning selection of bright extragalacticobjects to search for new gravitationally lensed quasars 06/2019 - applying CatBoost to great performance
• Gaussian-mixture-model-based cluster analysis ofgamma-ray bursts in the BATSE catalogue 06/2019
• Anomaly Detection in the Open Supernova Catalog 05/2019 - using Isolation Tree. GP parameters are 9 fitted parameters of the Gaussianprocess kernel and the log-likelihood of the fit
• Large-Scale Statistical Survey of Magnetopause Reconnection 05/2019 - using a boosted regression tree
• [Machine learning reveals systematic accumulation of electric current in lead-up to solar flares 05/2019] - using SVM for solar flare forecasts
• Multiwavelength cluster mass estimates and machine learning 05/2019 - training on semi-analytical simulation data of galaxy cluster merger trees, using RF and boosted tree
• AUTOREGRESSIVE PLANET SEARCH: APPLICATION TO THE KEPLERMISSION 05/2019 - ARIMA and ARFIMA models for detrending and whitening the time series; combined with RF
• A Bayesian direct method implementation to fit emissionline spectra: Application to the primordial He abundance determination 05/2019 - using PyMC3
• [Comparison of Observed Galaxy Properties with Semianalytic Model Predictions using Machine Learning 05/2019] - just a three layer NN, so not very deep ;-)
• Exploring helical dynamos with machine learning 05/2019 - good work becauase of factor analysis and test of different ML methods: findings are that simple linearly indipendent measures/features do the trick for parameter inference.
• Identification of Young Stellar Object candidates in theGaiaDR2 x AllWISE catalogue with machine learning methods 05/2019 - accesing NN, SVM, k-NN and random forest and selected and pre-reduced features, including WISE data
• Identification of RR Lyrae stars in multiband, sparsely-sampled data from the Dark EnergySurvey using template fitting and Random Forest classification 05/2019
• TiK-means: Transformation-infusedK-meansclustering for skewed groups 04/2019 - cool!
• Shaping Asteroid Models Using Genetic Evolution (SAGE) 04/2019- asteroid modelling algorithm based solely on photometric lightcurve data and performs quite great also with concave shapes and self-shadowing
• Investigating the dark matter signal in the cosmic ray antiprotonflux with the machine learning method 03/2019 - trained a machine to predict the likelyhood functions for model-selection bootstrapping.
• Random Forest identification of the thin disk, thick disk and hal oGaia-DR2 white dwarf population 03/2019
• Efficient Selection of Quasar Candidates Based on Optical and Infrared Photometric Data Using Machine Learning 03/2019 - XGBoost and SVM classifiers
• A Machine Learning Artificial Neural Network Calibration of the Strong-Line Oxygen Abundance 03/2019 - tested several NN structures, best was one hidden layer
• Fast likelihood-free cosmology with neural densityestimators and active learning 03/2019 - A seminal paper for parameter inference of cosmological models. Great.
• Constraining the Thermal Properties of Planetary Surfaces using Machine Learning:Application to Airless Bodies 02/2019 - using a shallow NN to predict obervables to speed up MCMC
• AGN selection in the AKARI NEP deep fieldwith the fuzzy SVM algorithm 02/2019
• Can a machine learn the outcome of planetary collisions? - investigate three ML methods to predict the outcome of colission simulations
• Stellar Formation Rates for photometric samples ofgalaxies using machine learning methods. 02/2019 promising results through the combination of different ML techniques
• Machine Learning for the Zwicky Transient Facility 02/2019 - formulated goals to use DL in future , except of that gave an overview of ML and some implementations of them
• Towards Machine-assisted Meta-Studies: The HubbleConstant 02/2019 - utilizing rulebased (no DL) model for this Meta-study as well as ANN clasifier to clasify novel measures
• Clustering clusters: unsupervised machine learning on globular cluster structural parameters - Use the Partitioning around Mediods method on 100 gloabular lcuster sof the Milky-Way and find 2-3 clusters within the sample.
• AUTOREGRESSIVE PLANET SEARCH: METHODOLOGY - features from autoregressive algorithm on planet lightcurve are fed in RF classifier. Good results.
• A machine learning approach for identification and classification of symbiotic stars using 2MASS and WISE 01/2019 - with a decision tree
• DeepCool: Fast and Accurate Estimation of Cooling Rates in Irradiated Gas with Artificial Neural Networks 01/2019 - a great example of how to create fast and precise sub-models for cosmological simulations based on ML regression of training-simulations
• Accurate Identification of Galaxy Mergers with Imaging 01/2019 - Linear Discriminant Analysis
• Star formation rates and stellar masses from machine learning 01/2019 - using RF from scikit-learn
• Machine Learning on Difference Image Analysis: A comparison of methods for transient detection 12/2018 - wrote an open python lirary for transient detection
• Systematic Serendipity: A Test of Unsupervised Machine Learning as a Method for Anomaly Detection 12/2018 - succesfull at detecting outliers via 60 crafted numerical features and t-SNE
• Finding the origin of noise transients in LIGO data with machine learning 12/2018 - building noise models with decision trees and genetic algorithms
• Classification of Multiwavelength Transients with Machine Learning - random forest on wavelet decomposition for iamges
• Probabilistic Random Forest: A machine learning algorithm for noisy datasets - a probabilitstic RF that performs well with label or feature noise
• ROMAN: Reduced-Order Modeling with Artificial Neurons
• Classification of gravitational-wave glitches via dictionary learning
• MACHINE LEARNING APPLIED TO STAR-GALAXY-QSO CLASSIFICATION AND STELLAR EFFECTIVE TEMPERATURE REGRESSION 11/2018
• Towards a radially-resolved semi-analytic model for the evolution of disc galaxies tuned with machine learning 10/2018 - parameter inference with Neural Net Emulator, trained on simulations, validated by MCMC
• Classifying Lensed Gravitational Waves in the Geometrical Optics Limit with Machine Learning 10/2018 - a comparative analysis of three ML methods
• Scalable Solutions for Automated Single Pulse Identification and Classification in Radio Astronomy 10/2018 - using Random Forest (XGBoost?)
• TSARDI: a Machine Learning data rejection algorithm for transiting exoplanet light curves 09/2018 - using DBSCAN
• Bayesian sparse reconstruction: a brute-force approach to astronomical imaging and machine learning 09/2018
• From FATS to feets: Further improvements to an astronomical feature extraction tool based on machine learning
• Protostellar classification using supervised machine learning algorithms 08/2018 - a comparison of different classifiers
• Machine Learning Classification of Gaia Data Release 2 08/2018 - with random forest
• Single-pulse classifier for the LOFAR Tied-Array All-sky Survey 08/2018 - using the WEKA tool and a fast decision tree for imbalanced problems
• A novel single-pulse search approach to detection of dispersed radio pulses using clustering and supervised machine learning
• A MACHINE-LEARNING METHOD FOR IDENTIFYING MULTI-WAVELENGTH COUNTERPARTS OF SUBMILLIMETER GALAXIES: TRAINING AND TESTING USING AS2UDS AND ALESS 2018
• Supervised Machine Learning for Analysing Spectra of Exoplanetary Atmospheres 2018
• Habitability Classification of Exoplanets: A Machine Learning Insight 2018
• Return of the features - Efficient feature selection and interpretation for photometric redshifts *2018
• Dissecting stellar chemical abundance space with t-SNE 2018
• GAME: GAlaxy Machine learning for Emission lines 2018
• INTEGRATING HUMAN AND MACHINE INTELLIGENCE IN GALAXY MORPHOLOGY CLASSIFICATION TASKS 2018
• Estimating Photometric Redshifts for X-ray sources in the X-ATLAS field, using machine-learning techniques 2017
• Learning from the machine: interpreting machine learning algorithms for point- and extended- source classification

Many groups now feel the urge to develop more into the ML/DL direction. Some publications advocate the use of DL

• Foreword to the Focus Issue on Machine Learning in Astronomy and Astrophysics 06/2019
• Machine learning and the physical sciences 03/2019 - seems like a great summary paper
• The Promise of Data Science for the Technosignatures Field 03/2019
• Astro2020 Science White Paper: The Role of Machine Learning in the NextDecade of Cosmology 02/2019
• Pushing the Technical Frontier: From Overwhelmingly Large Data Sets to Machine Learning 01/2019
• A high-bias, low-variance introduction to Machine Learning for physicists 03/2018

It is quite indicative that there are now working groups establishing to focus on machine learning & astrophysics like BABL AI, SkyML, Heidelberg AIN, and space.ml. In addition to that astrophysical data has not the same data security concerns like e.g. medicine or social science.