arXiv subjects

James E. Geach

Publications and source records attributed to James E. Geach.

At least 19 recordsLinked to original sources

arXiv2025

JWST Discovery of Warm Dust in the Circumgalactic Medium of the Makani Galaxy

We report the detection of near- and mid-infrared emission from polycyclic aromatic hydrocarbons (PAHs) out to ~ 35 kpc in the Makani Galaxy, a compact massive galaxy with a record-breaking 100-kpc scale starburst-driven wind at redshift z = 0.459. The NIRCam and MIRI observations with JWST take advantage of a coincidental match between the PAH spectral features at 3.3, 7.7, and (11.3 + 12.2) microns in Makani and the bandpasses of the MIRI and NIRCam filters. The warm dust is not only detected in the cool-gas tracers of the galactic wind associated with the more recent (7 Myr) starburst episode, but also in the outer warm-ionized gas wind produced by the older (0.4 Gyr) episode. The presence of PAHs in the outer wind indicates that the PAHs have survived the long (R/v ~ 10^8 yrs) journey to the halo despite the harsh environment of the galactic wind. The measured F1800W/F1130W flux ratios in the unresolved nucleus, inner halo (R = 10 - 20 kpc), and outer halo (R = 20 - 35 kpc), tracers of the PAH (11.3 + 12.2)/7.7 ratios, indicate decreasing starlight intensity incident on the PAHs, decreasing PAH sizes, and increasing PAH ionization fractions with increasing distance from the nucleus. These data provide the strongest evidence to date that the ejected dust of galactic winds survives the long journey to the CGM, but is eroded along the way.

astro-ph.GA↗

arXiv2025

Dargana: fine-tuning EarthPT for dynamic tree canopy mapping from space

We present Dargana, a fine-tuned variant of the EarthPT time-series foundation model that achieves specialisation using <3% of its pre-training data volume and 5% of its pre-training compute. Dargana is fine-tuned to generate regularly updated classification of tree canopy cover at 10m resolution, distinguishing conifer and broadleaved tree types. Using Cornwall, UK, as a test case, the model achieves a pixel-level ROC-AUC of 0.98 and a PR-AUC of 0.83 on unseen satellite imagery. Dargana can identify fine structures like hedgerows and coppice below the training sample limit, and can track temporal changes to canopy cover such as new woodland establishment. Our results demonstrate how pre-trained Large Observation Models like EarthPT can be specialised for granular, dynamic land cover monitoring from space, providing a valuable, scalable tool for natural capital management and conservation.

physics.geo-ph↗

arXiv2024

A kiloparsec-scale ordered magnetic field in a galaxy at z=5.6

Magnetic fields are widely observed in various astronomical contexts, yet much remains unknown about their significance across different systems and cosmic epochs. Our current knowledge of the evolution of magnetic fields is limited by scarce observations in the distant Universe, where galaxies have recently been found to be more evolved than most model predictions. To address this gap, we conducted rest-frame 131 um full-polarisation observations of dust emission in a strongly lensed dusty star-forming galaxy, SPT0346-52, at z=5.6, when the Universe was only 1 Gyr old. Dust grains can become aligned with local magnetic fields, resulting in the emission of linearly polarised thermal infrared radiation. Our observations have revealed a median polarisation level of $0.9\pm0.2$ percent with a variation of $\pm0.4$ percent across the 3 kpc extention, indicating the presence of large-scale ordered magnetic fields. The polarised dust emission is patchy, offset from the total dust emission and mostly overlaps with the [C II] emission at a velocity of about -150 km/s. The bimodal distribution of field orientations, their spatial distribution, and the connection with the cold gas kinematics further emphasise the complexity of the magnetic environment in this galaxy and the potential role of mergers in shaping its magnetic fields. Such early formation of ordered galactic magnetic fields also suggests that both small-scale and large-scale dynamos could be efficient in early galaxies. Continued observations of magnetic fields in early galaxies, as well as expanding surveys to a wider galaxy population, are essential for a comprehensive understanding of the prevalence and impact of magnetic fields in the evolving Universe.

astro-ph.GA↗

arXiv2024

The Outflowing [OII] Nebulae of Compact Starburst Galaxies at z $\sim$ 0.5

High-velocity outflows are ubiquitous in compact, massive (M$_* \sim$ 10$^{11}$ M$_{\odot}$), z $\sim$ 0.5 galaxies with extreme star formation surface densities ($Σ_{SFR} \sim$ 2000 M$_{\odot}$ yr$^{-1}$ kpc$^{-2}$). We have previously detected and characterized these outflows using MgII absorption lines. To probe their full extent, we present Keck/KCWI integral field spectroscopy of the [OII] and MgII emission nebulae surrounding all of the 12 galaxies in this study. We find that [OII] is more effective than MgII in tracing low surface brightness, extended emission in these galaxies. The [OII] nebulae are spatially extended beyond the stars, with radial extent R$_{90}$ between 10 and 40 kpc. The nebulae exhibit non-gravitational motions, indicating galactic outflows with maximum blueshifted velocities ranging from -335 to -1920 km s$^{-1}$. The outflow kinematics correlate with the bursty star formation histories of these galaxies. Galaxies with the most recent bursts of star formation (within the last $<$ 3 Myr) exhibit the highest central velocity dispersions ($σ>$ 400 km s$^{-1}$), while the oldest bursts have the lowest-velocity outflows. Many galaxies exhibit both high-velocity cores and more extended, slower-moving gas indicative of multiple outflow episodes. The slower, larger outflows occurred earlier and have decelerated as they propagate into the CGM and mix on timescales $>$ 50 Myr.

astro-ph.GA↗

arXiv2024

EarthPT: a time series foundation model for Earth Observation

We introduce EarthPT -- an Earth Observation (EO) pretrained transformer. EarthPT is a 700 million parameter decoding transformer foundation model trained in an autoregressive self-supervised manner and developed specifically with EO use-cases in mind. We demonstrate that EarthPT is an effective forecaster that can accurately predict future pixel-level surface reflectances across the 400-2300 nm range well into the future. For example, forecasts of the evolution of the Normalised Difference Vegetation Index (NDVI) have a typical error of approximately 0.05 (over a natural range of -1 -> 1) at the pixel level over a five month test set horizon, out-performing simple phase-folded models based on historical averaging. We also demonstrate that embeddings learnt by EarthPT hold semantically meaningful information and could be exploited for downstream tasks such as highly granular, dynamic land use classification. Excitingly, we note that the abundance of EO data provides us with -- in theory -- quadrillions of training tokens. Therefore, if we assume that EarthPT follows neural scaling laws akin to those derived for Large Language Models (LLMs), there is currently no data-imposed limit to scaling EarthPT and other similar `Large Observation Models.'

cs.LG↗

arXiv2023

Astronomia ex machina: a history, primer, and outlook on neural networks in astronomy

In this review, we explore the historical development and future prospects of artificial intelligence (AI) and deep learning in astronomy. We trace the evolution of connectionism in astronomy through its three waves, from the early use of multilayer perceptrons, to the rise of convolutional and recurrent neural networks, and finally to the current era of unsupervised and generative deep learning methods. With the exponential growth of astronomical data, deep learning techniques offer an unprecedented opportunity to uncover valuable insights and tackle previously intractable problems. As we enter the anticipated fourth wave of astronomical connectionism, we argue for the adoption of GPT-like foundation models fine-tuned for astronomical applications. Such models could harness the wealth of high-quality, multimodal astronomical data to serve state-of-the-art downstream tasks. To keep pace with advancements driven by Big Tech, we propose a collaborative, open-source approach within the astronomy community to develop and maintain these foundation models, fostering a symbiotic relationship between AI and astronomy that capitalizes on the unique strengths of both fields.

astro-ph.IM↗

arXiv2023

Kinematics, Structure, and Mass Outflow Rates of Extreme Starburst Galactic Outflows

We present results on the properties of extreme gas outflows in massive ($\rm M_* \sim$10$^{11} \ \rm M_{\odot}$), compact, starburst ($\rm SFR \sim$$200 \, \rm M_{\odot} \ yr^{-1}$) galaxies at z = $0.4-0.7$ with very high star formation surface densities ($\rm Σ_{SFR} \sim$$2000 \,\rm M_{\odot} \ yr^{-1} \ kpc^{-2}$). Using optical Keck/HIRES spectroscopy of 14 HizEA starburst galaxies we identify outflows with maximum velocities of $820 - 2860$ \kmps. High-resolution spectroscopy allows us to measure precise column densities and covering fractions as a function of outflow velocity and characterize the kinematics and structure of the cool gas outflow phase (T $\sim$10$^4$ K). We find substantial variation in the absorption profiles, which likely reflects the complex morphology of inhomogeneously-distributed, clumpy gas and the intricacy of the turbulent mixing layers between the cold and hot outflow phases. There is not a straightforward correlation between the bursts in the galaxies' star formation histories and their wind absorption line profiles, as might naively be expected for starburst-driven winds. The lack of strong \mgii \ absorption at the systemic velocity is likely an orientation effect, where the observations are down the axis of a blowout. We infer high mass outflow rates of $\rm \sim$50 $-$ 2200 $\rm M_{\odot} \, yr^{-1}$, assuming a fiducial outflow size of 5 kpc, and mass loading factors of $η\sim$5 for most of the sample. %with $η\sim$20 for two galaxies. While these values have high uncertainties, they suggest that starburst galaxies are capable of ejecting very large amounts of cool gas that will substantially impact their future evolution.

astro-ph.GA↗

arXiv2023

The Ionization and Dynamics of the Makani Galactic Wind

The Makani galaxy hosts the poster child of a galactic wind on scales of the circumgalactic medium. It consists of a two-episode wind in which the slow, outer wind originated 400 Myr ago (Episode I; R_I = 20-50 kpc) and the fast, inner wind is 7 Myr old (Episode II; R_II = 0-20 kpc). While this wind contains ionized, neutral, and molecular gas, the physical state and mass of the most extended phase--the warm, ionized gas--is unknown. Here we present Keck optical spectra of the Makani outflow. These allow us to detect hydrogen lines out to r = 30-40 kpc and thus constrain the mass, momentum, and energy in the wind. Many collisionally-excited lines are detected throughout the wind, and their line ratios are consistent with 200-400 km/s shocks that power the ionized gas, with v_shock = $σ$_wind. Combining shock models, density-sensitive line ratios, and mass and velocity measurements, we estimate that the ionized mass and outflow rate in the Episode II wind could be as high as that of the molecular gas: M_II(HII) ~ M_II(H_2) = (1-2)x10^9 Msun and dM/dt_II(HII) ~ dM/dt_II(H_2) = 170-250 Msun/yr. The outer wind has slowed, so that dM/dt_I(HII) ~ 10 Msun/yr, but it contains more ionized gas: M_I(HII) = 5x10^9 Msun. The momentum and energy in the recent Episode II wind imply a momentum-driven flow (p ``boost" ~ 7) driven by the hot ejecta and radiation pressure from the Eddington-limited, compact starburst. Much of the energy and momentum in the older Episode I wind may reside in a hotter phase, or lie further into the CGM.

astro-ph.GA↗

arXiv2023

Host Dark Matter Halos of WISE-selected Obscured & Unobscured Quasars: Evidence for Evolution

Obscuration in quasars may arise from steep viewing angles along the dusty torus, or instead may represent a distinct phase of supermassive black hole growth. We test these scenarios by probing the host dark matter halo environments of $\sim 1.4$ million WISE-selected obscured and unobscured quasars at $\langle z \rangle = 1.4$ using angular clustering measurements as well as cross-correlation measurements of quasar positions with the gravitational lensing of the cosmic microwave background (CMB). We interpret these signals within a halo occupation distribution (HOD) framework to conclude that obscured systems reside in more massive effective halos ($ \sim 10^{12.9} h^{-1} M_{\odot}$) than their unobscured counterparts ($ \sim 10^{12.6} h^{-1} M_{\odot}$), though we do not detect a difference in the satellite fraction. We find excellent agreement between the clustering and lensing analyses and show that this implies the observed difference is robust to uncertainties in the obscured quasar redshift distribution, highlighting the power of combining angular clustering and weak lensing measurements. This finding appears in tension with models that ascribe obscuration exclusively to orientation of the dusty torus along the line-of-sight, and instead may be consistent with the notion that some obscured quasars are attenuated by galaxy-scale or circumnuclear material during an evolutionary phase.

astro-ph.GA↗

arXiv2022

The Space Density of Intermediate Redshift, Extremely Compact, Massive Starburst Galaxies

We present a measurement of the intrinsic space density of intermediate redshift ($z\sim0.5$), massive ($M_{*} \sim 10^{11} \ \text{M}_{\odot}$), compact ($R_{e} \sim 100$ pc) starburst ($Σ_{SFR} \sim 1000 \ \text{M}_{\odot} \ \text{yr}^{-1} \text{kpc}^{-1}$) galaxies with tidal features indicative of them having undergone recent major mergers. A subset of them host kiloparsec scale, $>1000 \ \text{km}\ \text{s}^{-1}$ outflows and have little indication of AGN activity, suggesting that extreme star formation can be a primary driver of large-scale feedback. The aim for this paper is to calculate their space density so we can place them in a better cosmological context. We do this by empirically modeling the stellar populations of massive, compact starburst galaxies. We determine the average timescale for which galaxies that have recently undergone an extreme nuclear starburst would be targeted and included in our spectroscopically selected sample. We find that massive, compact starburst galaxies targeted by our criteria would be selectable for $\sim 148 ^{+27}_{-24}$ Myr and have an intrinsic space density $n_{\text{CS}} \sim (1.1^{+0.5}_{-0.3}) \times 10^{-6} \ \ \text{Mpc}^{-3}$. This space density is broadly consistent with our $z\sim0.5$ compact starbursts being the most extremely compact and star forming low redshift analogs of the compact star forming galaxies in the early Universe as well as them being the progenitors to a fraction of intermediate redshift post starburst and compact quiescent galaxies.

astro-ph.GA↗

arXiv2022

Realistic galaxy image simulation via score-based generative models

We show that a Denoising Diffusion Probabalistic Model (DDPM), a class of score-based generative model, can be used to produce realistic mock images that mimic observations of galaxies. Our method is tested with Dark Energy Spectroscopic Instrument (DESI) grz imaging of galaxies from the Photometry and Rotation curve OBservations from Extragalactic Surveys (PROBES) sample and galaxies selected from the Sloan Digital Sky Survey. Subjectively, the generated galaxies are highly realistic when compared with samples from the real dataset. We quantify the similarity by borrowing from the deep generative learning literature, using the `Fréchet Inception Distance' to test for subjective and morphological similarity. We also introduce the `Synthetic Galaxy Distance' metric to compare the emergent physical properties (such as total magnitude, colour and half light radius) of a ground truth parent and synthesised child dataset. We argue that the DDPM approach produces sharper and more realistic images than other generative methods such as Adversarial Networks (with the downside of more costly inference), and could be used to produce large samples of synthetic observations tailored to a specific imaging survey. We demonstrate two potential uses of the DDPM: (1) accurate in-painting of occluded data, such as satellite trails, and (2) domain transfer, where new input images can be processed to mimic the properties of the DDPM training set. Here we `DESI-fy' cartoon images as a proof of concept for domain transfer. Finally, we suggest potential applications for score-based approaches that could motivate further research on this topic within the astronomical community.

astro-ph.IM↗

arXiv2022

Host Dark Matter Halos of SDSS Red and Blue Quasars: No Significant Difference in Large-scale Environment

The observed optical colors of quasars are generally interpreted in one of two frameworks: unified models which attribute color to random orientation of the accretion disk along the line-of-sight, and evolutionary models which invoke connections between quasar systems and their environments. We test these schema by probing the dark matter halo environments of optically-selected quasars as a function of $g-i$ optical color by measuring the two-point correlation functions of $\sim$ 0.34 million eBOSS quasars as well as the gravitational deflection of cosmic microwave background photons around $\sim$ 0.66 million XDQSO photometric quasar candidates. We do not detect a trend of halo bias with optical color through either analysis, finding that optically-selected quasars at $0.8 < z < 2.2$ occupy halos of characteristic mass $M_{h}\sim 3\times 10^{12} \ h^{-1} M_{\odot}$ regardless of their color. This result implies that a quasar's large-scale halo environment is not strongly connected to its observed optical color. We also confirm findings of fundamental differences in the radio properties of red and blue quasars by stacking 1.4 GHz FIRST images at their positions, suggesting the observed differences cannot be attributed to orientation. Instead, the differences between red and blue quasars likely arise on nuclear-galactic scales, perhaps owing to reddening by a nuclear dusty wind. Finally, we show that optically-selected quasars' halo environments are also independent of their $r-W2$ optical-infrared colors, while previous work has suggested that mid-infrared-selected obscured quasars occupy more massive halos. We discuss implications of this result for models of quasar and galaxy co-evolution.

astro-ph.GA↗

arXiv2021

Physical Properties of Massive Compact Starburst Galaxies with Extreme Outflows

We present results on the nature of extreme ejective feedback episodes and the physical conditions of a population of massive ($\rm M_* \sim 10^{11} M_{\odot}$), compact starburst galaxies at z = 0.4-0.7. We use data from Keck/NIRSPEC, SDSS, Gemini/GMOS, MMT, and Magellan/MagE to measure rest-frame optical and near-IR spectra of 14 starburst galaxies with extremely high star formation rate surface densities (mean $\rm Σ_{SFR} \sim 3000 \,M_{\odot} yr^{-1} kpc^{-2}$) and powerful galactic outflows (maximum speeds v$_{98} \sim$ 1000-3000 km s$^{-1}$). Our unique data set includes an ensemble of both emission [OII]$λλ$3726,3729, H$β$, [OIII]$λλ$4959,5007, H$α$, [NII]$λλ$6548,6583, and [SII]$λλ$6716,6731) and absorption MgII$λλ$2796,2803, and FeII$λ$2586) lines that allow us to investigate the kinematics of the cool gas phase (T$\sim$10$^4$ K) in the outflows. Employing a suite of line ratio diagnostic diagrams, we find that the central starbursts are characterized by high electron densities (median n$_e \sim$ 530 cm$^{-3}$), and high metallicity (solar or super-solar). We show that the outflows are most likely driven by stellar feedback emerging from the extreme central starburst, rather than by an AGN. We also present multiple intriguing observational signatures suggesting that these galaxies may have substantial Lyman continuum (LyC) photon leakage, including weak [SII] nebular emission lines. Our results imply that these galaxies may be captured in a short-lived phase of extreme star formation and feedback where much of their gas is violently blown out by powerful outflows that open up channels for LyC photons to escape.

astro-ph.GA↗

arXiv2021

Pix2Prof: fast extraction of sequential information from galaxy imagery via a deep natural language 'captioning' model

We present 'Pix2Prof', a deep learning model that can eliminate any manual steps taken when extracting galaxy profiles. We argue that a galaxy profile of any sort is conceptually similar to a natural language image caption. This idea allows us to leverage image captioning methods from the field of natural language processing, and so we design Pix2Prof as a float sequence 'captioning' model suitable for galaxy profile inference. We demonstrate the technique by approximating a galaxy surface brightness (SB) profile fitting method that contains several manual steps. Pix2Prof processes $\sim$1 image per second on an Intel Xeon E5 2650 v3 CPU, improving on the speed of the manual interactive method by more than two orders of magnitude. Crucially, Pix2Prof requires no manual interaction, and since galaxy profile estimation is an embarrassingly parallel problem, we can further increase the throughput by running many Pix2Prof instances simultaneously. In perspective, Pix2Prof would take under an hour to infer profiles for $10^5$ galaxies on a single NVIDIA DGX-2 system. A single human expert would take approximately two years to complete the same task. Automated methodology such as this will accelerate the analysis of the next generation of large area sky surveys expected to yield hundreds of millions of targets. In such instances, all manual approaches -- even those involving a large number of experts -- will be impractical.

astro-ph.IM↗

arXiv2021

Compact Starburst Galaxies with Fast Outflows: Central Escape Velocities and Stellar Mass Surface Densities from Multi-band Hubble Space Telescope Imaging

We present multi-band Hubble Space Telescope imaging that spans rest-frame near-ultraviolet through near-infrared wavelengths (0.3-1.1 $μ$m) for 12 compact starburst galaxies at z=0.4-0.8. These massive galaxies (M_stellar ~ 10^11 M_Sun) are driving very fast outflows ($v_{max}$=1000-3000 km/s), and their light profiles are dominated by an extremely compact starburst component (half-light radius ~ 100 pc). Our goal is to constrain the physical mechanisms responsible for launching these fast outflows by measuring the physical conditions within the central kiloparsec. Based on our stellar population analysis, the central component typically contributes $\approx$25% of the total stellar mass and the central escape velocities $v_{esc,central}\approx900$ km/s are a factor of two smaller than the observed outflow velocities. This requires physical mechanisms that can accelerate gas to speeds significantly beyond the central escape velocities, and it makes clear that these fast outflows are capable of traveling into the circumgalactic medium, and potentially beyond. We find central stellar densities comparable to theoretical estimates of the Eddington limit, and we estimate $Σ_1$ surface densities within the central kpc comparable to those of compact massive galaxies at $0.5<z<3.0$. Relative to "red nuggets" and "blue nuggets" at $z\sim2$, we find significantly smaller $r_e$ values at a given stellar mass, which we attribute to the dominance of a young stellar component in our sample and the better physical resolution for rest-frame optical observations at $z\sim0.6$ versus $z\sim2$. We compare to theoretical scenarios involving major mergers and violent disc instability, and we speculate that our galaxies are progenitors of power-law ellipticals in the local universe with prominent stellar cusps.

astro-ph.GA↗

arXiv2021

Rapid sorting of radio galaxy morphology using Haralick features

We demonstrate the use of Haralick features for the automated classification of radio galaxies. The set of thirteen Haralick features represent an extremely compact non-parametric representation of image texture, and are calculated directly from imagery using the Grey Level Co-occurrence Matrix (GLCM). The GLCM is an encoding of the relationship between the intensity of neighbouring pixels in an image. Using 10,000 sources detected in the first data release of the LOFAR Two-metre Sky Survey (LoTSS), we demonstrate that Haralick features are highly efficient, rotationally invariant descriptors of radio galaxy morphology. After calculating Haralick features for LoTSS sources, we employ the fast density-based hierarchical clustering algorithm HDBSCAN to group radio sources into a sequence of morphological classes, illustrating a simple methodology to classify and label new, unseen galaxies in large samples. By adopting a 'soft' clustering approach, we can assign each galaxy a probability of belonging to a given cluster, allowing for more flexibility in the selection of galaxies according to combinations of morphological characteristics and for easily identifying outliers: those objects with a low probability of belonging to any cluster in the Haralick space. Although our demonstration focuses on radio galaxies, Haralick features can be calculated for any image, making this approach also relevant to large optical imaging galaxy surveys.

astro-ph.IM↗

arXiv2021

Reproducing sub-millimetre galaxy number counts with cosmological hydrodynamic simulations

Matching the number counts of high-$z$ sub-millimetre-selected galaxies (SMGs) has been a long standing problem for galaxy formation models. In this paper, we use 3D dust radiative transfer to model the sub-mm emission from galaxies in the SIMBA cosmological hydrodynamic simulations, and compare predictions to the latest single-dish observational constraints on the abundance of 850$\mathrm{μm}$-selected sources. We find good agreement with the shape of the integrated 850$\mathrm{μm}$ luminosity function, and the normalisation is within 0.25 dex at $> 3 \; \mathrm{mJy}$, unprecedented for a fully cosmological hydrodynamic simulation, along with good agreement in the redshift distribution of bright SMGs. The agreement is driven primarily by SIMBA's good match to infrared measures of the star formation rate (SFR) function between $z = 2-4$ at high SFRs. Also important is the self-consistent on-the-fly dust model in SIMBA, which predicts, on average, higher dust masses (by up to a factor of 2.5) compared to using a fixed dust-to-metals ratio of 0.3. We construct a lightcone to investigate the effect of far-field blending, and find that 52% of sources are blends of multiple components, which makes a small contribution to the normalisation of the bright-end of the number counts. We provide new fits to the 850$\mathrm{μm}$ luminosity as a function of SFR and dust mass. Our results demonstrate that exotic solutions to the discrepancy between sub-mm counts in simulations and observations, such as a top-heavy IMF, are unnecessary, and that sub-millimetre-bright phases are a natural consequence of massive galaxy evolution.

astro-ph.GA↗

arXiv2020

AstroVaDEr: Astronomical Variational Deep Embedder for Unsupervised Morphological Classification of Galaxies and Synthetic Image Generation

We present AstroVaDEr, a variational autoencoder designed to perform unsupervised clustering and synthetic image generation using astronomical imaging catalogues. The model is a convolutional neural network that learns to embed images into a low dimensional latent space, and simultaneously optimises a Gaussian Mixture Model (GMM) on the embedded vectors to cluster the training data. By utilising variational inference, we are able to use the learned GMM as a statistical prior on the latent space to facilitate random sampling and generation of synthetic images. We demonstrate AstroVaDEr's capabilities by training it on gray-scaled \textit{gri} images from the Sloan Digital Sky Survey, using a sample of galaxies that are classified by Galaxy Zoo 2. An unsupervised clustering model is found which separates galaxies based on learned morphological features such as axis ratio, surface brightness profile, orientation and the presence of companions. We use the learned mixture model to generate synthetic images of galaxies based on the morphological profiles of the Gaussian components. AstroVaDEr succeeds in producing a morphological classification scheme from unlabelled data, but unexpectedly places high importance on the presence of companion objects---demonstrating the importance of human interpretation. The network is scalable and flexible, allowing for larger datasets to be classified, or different kinds of imaging data. We also demonstrate the generative properties of the model, which allow for realistic synthetic images of galaxies to be sampled from the learned classification scheme. These can be used to create synthetic image catalogs or to perform image processing tasks such as deblending.

astro-ph.IM↗