arXiv subjects

John F. Wu

Publications and source records attributed to John F. Wu.

At least 37 records · Page 2Linked to original sources

arXiv2024

Estimating Dark Matter Halo Masses in Simulated Galaxy Clusters with Graph Neural Networks

Galaxies grow and evolve in dark matter halos. Because dark matter is not visible, galaxies' halo masses ($\rm{M}_{\rm{halo}}$) must be inferred indirectly. We present a graph neural network (GNN) model for predicting $\rm{M}_{\rm{halo}}$ from stellar mass ($\rm{M}_{*}$) in simulated galaxy clusters using data from the IllustrisTNG simulation suite. Unlike traditional machine learning models like random forests, our GNN captures the information-rich substructure of galaxy clusters by using spatial and kinematic relationships between galaxy neighbour. A GNN model trained on the TNG-Cluster dataset and independently tested on the TNG300 simulation achieves superior predictive performance compared to other baseline models we tested. Future work will extend this approach to different simulations and real observational datasets to further validate the GNN model's ability to generalise.

astro-ph.GA↗

arXiv2024

ALMA Lensing Cluster Survey: Dust mass measurements as a function of redshift, stellar-mass and star formation rate, from z=1 to z=5

Understanding the dust content of galaxies, its evolution with redshift and its relationship to stars and star formation is fundamental for our understanding of galaxy evolution. Using the ALMA Lensing Cluster Survey (ALCS) wide-area band-6 continuum dataset ($\sim\,$110 arcmin$^2$ across 33 lensing clusters), we aimed at constraining the dust mass evolution with redshift, stellar mass and star formation rate (SFR). After binning sources according to redshift, SFR and stellar mass -- extracted from an HST-IRAC catalog -- we performed a set of continuum stacking analyses in the image domain using \textsc{LineStacker} on sources between $z=1$ and $z=5$, further improving the depth of our data. The large field of view provided by the ALCS allows us to reach a final sample of $\sim4000$ galaxies with known coordinates and SED-derived physical parameters. We stack sources with SFR between $10^{-3}$ and $10^{3}$ M$_\odot$ per year, and stellar mass between $10^{8}$ and $10^{12}$ M$_\odot$, splitting them in different stellar mass and SFR bins. Through stacking we retrieve the continuum 1.2\,mm flux, a known dust mass tracer, allowing us to derive the dust mass evolution with redshift and its relation with SFR and stellar mass. We observe clear continuum detections in the majority of the subsamples. From the non detections we derive 3-$σ$ upper limits. We observe a steady decline in the average dust mass with redshift. Moreover, sources with higher stellar mass or SFR have higher dust mass on average, allowing us to derive scaling relations. Our results are mostly in good agreement with models at $z\sim1$-3, but indicate typically lower dust-mass than predicted at higher redshift.

astro-ph.GA↗

arXiv2024

The SAGA Survey. III. A Census of 101 Satellite Systems around Milky Way-mass Galaxies

We present Data Release 3 (DR3) of the Satellites Around Galactic Analogs (SAGA) Survey, a spectroscopic survey characterizing satellite galaxies around Milky Way (MW)-mass galaxies. The SAGA Survey DR3 includes 378 satellites identified across 101 MW-mass systems in the distance range of 25-40.75 Mpc, and an accompanying redshift catalog of background galaxies (including about 46,000 taken by SAGA) in the SAGA footprint of 84.7 sq. deg. The number of confirmed satellites per system ranges from zero to 13, in the stellar mass range of 10^6 to 10^10 solar masses. Based on a detailed completeness model, this sample accounts for 94% of the true satellite population down to a stellar mass of 10^7.5 solar masses. We find that the mass of the most massive satellite in SAGA systems is the strongest predictor of satellite abundance; one-third of the SAGA systems contain LMC-mass satellites, and they tend to have more satellites than the MW. The SAGA satellite radial distribution is less concentrated than the MW's, and the SAGA quenched fraction below 10^8.5 solar masses is lower than the MW's, but in both cases, the MW is within 1 sigma of SAGA system-to-system scatter. SAGA satellites do not exhibit a clear corotating signal as has been suggested in the MW/M31 satellite systems. Although the MW differs in many respects from the typical SAGA system, these differences can be reconciled if the MW is an older, slightly less massive host with a recently accreted LMC/SMC system.

astro-ph.GA↗

arXiv2024

How the Galaxy-Halo Connection Depends on Large-Scale Environment

We investigate the connection between galaxies, dark matter halos, and their large-scale environments at $z=0$ with Illustris TNG300 hydrodynamic simulation data. We predict stellar masses from subhalo properties to test two types of machine learning (ML) models: Explainable Boosting Machines (EBMs) with simple galaxy environment features and $\mathbb{E}(3)$-invariant graph neural networks (GNNs). The best-performing EBM models leverage spherically averaged overdensity features on $3$ Mpc scales. Interpretations via SHapley Additive exPlanations (SHAP) also suggest that, in the context of the TNG300 galaxy-halo connection, simple spherical overdensity on $\sim 3$ Mpc scales is more important than cosmic web distance features measured using the DisPerSE algorithm. Meanwhile, a GNN with connectivity defined by a fixed linking length, $L$, outperforms the EBM models by a significant margin. As we increase the linking length scale, GNNs learn important environmental contributions up to the largest scales we probe ($L=10$ Mpc). We conclude that $3$ Mpc distance scales are most critical for describing the TNG galaxy-halo connection using the spherical overdensity parameterization but that information on larger scales, which is not captured by simple environmental parameters or cosmic web features, can further augment these models. Our study highlights the benefits of using interpretable ML algorithms to explain models of astrophysical phenomena, and the power of using GNNs to flexibly learn complex relationships directly from data while imposing constraints from physical symmetries.

astro-ph.GA↗

arXiv2024

Disentangling Dense Embeddings with Sparse Autoencoders

Sparse autoencoders (SAEs) have shown promise in extracting interpretable features from complex neural networks. We present one of the first applications of SAEs to dense text embeddings from large language models, demonstrating their effectiveness in disentangling semantic concepts. By training SAEs on embeddings of over 420,000 scientific paper abstracts from computer science and astronomy, we show that the resulting sparse representations maintain semantic fidelity while offering interpretability. We analyse these learned features, exploring their behaviour across different model capacities and introducing a novel method for identifying ``feature families'' that represent related concepts at varying levels of abstraction. To demonstrate the practical utility of our approach, we show how these interpretable features can be used to precisely steer semantic search, allowing for fine-grained control over query semantics. This work bridges the gap between the semantic richness of dense embeddings and the interpretability of sparse representations. We open source our embeddings, trained sparse autoencoders, and interpreted features, as well as a web app for exploring them.

cs.LG↗

arXiv2024

pathfinder: A Semantic Framework for Literature Review and Knowledge Discovery in Astronomy

The exponential growth of astronomical literature poses significant challenges for researchers navigating and synthesizing general insights or even domain-specific knowledge. We present Pathfinder, a machine learning framework designed to enable literature review and knowledge discovery in astronomy, focusing on semantic searching with natural language instead of syntactic searches with keywords. Utilizing state-of-the-art large language models (LLMs) and a corpus of 350,000 peer-reviewed papers from the Astrophysics Data System (ADS), Pathfinder offers an innovative approach to scientific inquiry and literature exploration. Our framework couples advanced retrieval techniques with LLM-based synthesis to search astronomical literature by semantic context as a complement to currently existing methods that use keywords or citation graphs. It addresses complexities of jargon, named entities, and temporal aspects through time-based and citation-based weighting schemes. We demonstrate the tool's versatility through case studies, showcasing its application in various research scenarios. The system's performance is evaluated using custom benchmarks, including single-paper and multi-paper tasks. Beyond literature review, Pathfinder offers unique capabilities for reformatting answers in ways that are accessible to various audiences (e.g. in a different language or as simplified text), visualizing research landscapes, and tracking the impact of observatories and methodologies. This tool represents a significant advancement in applying AI to astronomical research, aiding researchers at all career stages in navigating modern astronomy literature.

astro-ph.IM↗

arXiv2024

The SAGA Survey. IV. The Star Formation Properties of 101 Satellite Systems around Milky Way-mass Galaxies

We present the star-forming properties of 378 satellite galaxies around 101 Milky Way analogs in the Satellites Around Galactic Analogs (SAGA) Survey, focusing on the environmental processes that suppress or quench star formation. In the SAGA stellar mass range of 10^6 to 10^10 solar masses, we present quenched fractions, star-forming rates, gas-phase metallicities, and gas content. The fraction of SAGA satellites that are quenched increases with decreasing stellar mass and shows significant system-to-system scatter. SAGA satellite quenched fractions are highest in the central 100 kpc of their hosts and decline out to the virial radius. Splitting by specific star formation rate (sSFR), the least star-forming satellite quartile follows the radial trend of the quenched population. The median sSFR of star-forming satellites increases with decreasing stellar mass and is roughly constant with projected radius. Star-forming SAGA satellites are consistent with the star formation rate--stellar mass relationship determined in the Local Volume, while the median gas-phase metallicity is higher and median HI gas mass is lower at all stellar masses. We investigate the dependence of the satellite quenched fraction on host properties. Quenched fractions are higher in systems with larger host halo mass, but this trend is only seen in the inner 100 kpc; we do not see significant trends with host color or star formation rate. Our results suggest that lower mass satellites and satellites inside 100 kpc are more efficiently quenched in a Milky Way-like environment, with these processes acting sufficiently slowly to preserve a population of star-forming satellites at all stellar masses and projected radii.

astro-ph.GA↗

arXiv2024

Predicting dark matter halo masses from simulated galaxy images and environments

Galaxies are theorized to form and co-evolve with their dark matter halos, such that their stellar masses and halo masses should be well-correlated. However, it is not known whether other observable galaxy features, such as their morphologies or large-scale environments, can be used to tighten the correlation between galaxy properties and halo masses. In this work, we train a baseline random forest model to predict halo mass using galaxy features from the Illustris TNG50 hydrodynamical simulation, and compare with convolutional neural networks (CNNs) and graph neural networks (GNNs) trained respectively using galaxy image cutouts and galaxy point clouds. The best baseline model has a root mean squared error (RMSE) of 0.310 and mean absolute error (MAE) of 0.220, compared to the CNN (RSME=0.359, MAE=0.238), GNN (RMSE=0.248, MAE=0.158), and a novel combined CNN+GNN (RMSE=0.248, MAE=0.144). The CNN is likely limited by our small data set, and we anticipate that the CNN and CNN+GNN would benefit from training on larger cosmological simulations. We conclude that deep learning models can leverage information from galaxy appearances and environment, beyond commonly used summary statistics, in order to better predict the halo mass.

astro-ph.CO↗

arXiv2024

Designing an Evaluation Framework for Large Language Models in Astronomy Research

Large Language Models (LLMs) are shifting how scientific research is done. It is imperative to understand how researchers interact with these models and how scientific sub-communities like astronomy might benefit from them. However, there is currently no standard for evaluating the use of LLMs in astronomy. Therefore, we present the experimental design for an evaluation study on how astronomy researchers interact with LLMs. We deploy a Slack chatbot that can answer queries from users via Retrieval-Augmented Generation (RAG); these responses are grounded in astronomy papers from arXiv. We record and anonymize user questions and chatbot answers, user upvotes and downvotes to LLM responses, user feedback to the LLM, and retrieved documents and similarity scores with the query. Our data collection method will enable future dynamic evaluations of LLM tools for astronomy.

astro-ph.IM↗

arXiv2024

Katachi: Decoding the Imprints of Past Star Formation on Present Day Morphology in Galaxies with Interpretable CNNs

The physical processes responsible for shaping how galaxies form and quench over time leave imprints on both the spatial (galaxy morphology) and temporal (star formation history; SFH) tracers that we use to study galaxies. While the morphology-SFR connection is well studied, the correlation with past star formation activity is not as well understood. To quantify this we present Katachi, an interpretable convolutional neural network (CNN) framework that learns the connection between the factors regulating star formation in galaxies on different spatial and temporal scales. Katachi is trained on 9904 galaxies at 0.02$<$z$<$0.1 in the SDSS-IV MaNGA DR17 sample to predict stellar mass (M$_*$; RMSE 0.22 dex), current star formation rate (SFR; RMSE 0.31 dex) and half-mass time (t$_{50}$; RMSE 0.23 dex). This information allows us to reconstruct non-parametric SFHs for each galaxy from \textit{gri} imaging alone. To quantify the morphological features informing the SFH predictions we use SHAP (SHapley Additive exPlanations). We recover the expected trends of M$_*$ governed by the growth of galaxy bulges, and SFR correlating with spiral arms and other star-forming regions. We also find the SHAP maps of D4000 are more complex than those of M$_*$ and SFR, and that morphology is correlated with t$_{50}$ even at fixed mass and SFR. Katachi serves as a scalable public framework to predict galaxy properties from large imaging surveys including Rubin, Roman, and Euclid, with large datasets of high SNR imaging across limited photometric bands.

astro-ph.GA↗

arXiv2024

PHANGS-ML: dissecting multiphase gas and dust in nearby galaxies using machine learning

The PHANGS survey uses ALMA, HST, VLT, and JWST to obtain an unprecedented high-resolution view of nearby galaxies, covering millions of spatially independent regions. The high dimensionality of such a diverse multi-wavelength dataset makes it challenging to identify new trends, particularly when they connect observables from different wavelengths. Here we use unsupervised machine learning algorithms to mine this information-rich dataset to identify novel patterns. We focus on three of the PHANGS-JWST galaxies, for which we extract properties pertaining to their stellar populations; warm ionized and cold molecular gas; and Polycyclic Aromatic Hydrocarbons (PAHs), as measured over 150 pc-scale regions. We show that we can divide the regions into groups with distinct multiphase gas and PAH properties. In the process, we identify previously-unknown galaxy-wide correlations between PAH band and optical line ratios and use our identified groups to interpret them. The correlations we measure can be naturally explained in a scenario where the PAHs and the ionized gas are exposed to different parts of the same radiation field that varies spatially across the galaxies. This scenario has several implications for nearby galaxies: (i) The uniform PAH ionized fraction on 150 pc scales suggests significant self-regulation in the ISM, (ii) the PAH 11.3/7.7 \mic~ band ratio may be used to constrain the shape of the non-ionizing far-ultraviolet to optical part of the radiation field, and (iii) the varying radiation field affects line ratios that are commonly used as PAH size diagnostics. Neglecting this effect leads to incorrect or biased PAH sizes.

astro-ph.GA↗

arXiv2023

JWST constraints on the UV luminosity density at cosmic dawn: implications for 21-cm cosmology

An unprecedented array of new observational capabilities are starting to yield key constraints on models of the epoch of first light in the Universe. In this Letter we discuss the implications of the UV radiation background at cosmic dawn inferred by recent JWST observations for radio experiments aimed at detecting the redshifted 21-cm hyperfine transition of diffuse neutral hydrogen. Under the basic assumption that the 21-cm signal is activated by the Ly$α$ photon field produced by metal-poor stellar systems, we show that a detection at the low frequencies of the EDGES and SARAS3 experiments may be expected from a simple extrapolation of the declining UV luminosity density inferred at $z\lesssim 14$ from JWST early galaxy data. Accounting for an early radiation excess above the CMB suggests a shallower or flat evolution to simultaneously reproduce low and high-$z$ current UV luminosity density constraints, which cannot be entirely ruled out, given the large uncertainties from cosmic variance and the faint-end slope of the galaxy luminosity function at cosmic dawn. Our findings raise the intriguing possibility that a high star formation efficiency at early times may trigger the onset of intense Ly$α$ emission at redshift $z\lesssim 20$ and produce a cosmic 21-cm absorption signal 200 Myr after the Big Bang.

astro-ph.CO↗

arXiv2023

Quantifying Roman WFI Dark Images with the Wavelet Scattering Transform

The Nancy Grace Roman Space Telescope will survey a large area of the sky at near-infrared wavelengths with its Wide Field Instrument (WFI). The performance of the 18 WFI H4RG-10 detectors will need to be well-characterized and regularly monitored in order for Roman to meet its science objectives. Weak lensing science goals are particularly sensitive to instrumental distortions and patterns that might masquerade as astronomical signals. We apply the wavelet scattering transform in order to analyze localized signals in Roman WFI images that have been taken as part of a dark image test suite. The scattering transform quantifies shapes and clustering information by reducing images into non-linear combinations of wavelet modes on multiple size scales. We show that these interpretable scattering statistics can separate rare correlated patterns from typical noise signals, and we discuss the results in context of power spectrum analyses and other computer vision methods.

astro-ph.IM↗

arXiv2023

Learning the galaxy-environment connection with graph neural networks

Galaxies co-evolve with their host dark matter halos. Models of the galaxy-halo connection, calibrated using cosmological hydrodynamic simulations, can be used to populate dark matter halo catalogs with galaxies. We present a new method for inferring baryonic properties from dark matter subhalo properties using message-passing graph neural networks (GNNs). After training on subhalo catalog data from the Illustris TNG300-1 hydrodynamic simulation, our GNN can infer stellar mass from the host and neighboring subhalo positions, kinematics, masses, and maximum circular velocities. We find that GNNs can also robustly estimate stellar mass from subhalo properties in 2d projection. While other methods typically model the galaxy-halo connection in isolation, our GNN incorporates information from galaxy environments, leading to more accurate stellar mass inference.

astro-ph.IM↗

arXiv2023

Target Selection and Sample Characterization for the DESI LOW-Z Secondary Target Program

We introduce the DESI LOW-Z Secondary Target Survey, which combines the wide-area capabilities of the Dark Energy Spectroscopic Instrument (DESI) with an efficient, low-redshift target selection method. Our selection consists of a set of color and surface brightness cuts, combined with modern machine learning methods, to target low-redshift dwarf galaxies ($z$ < 0.03) between $19 < r < 21$ with high completeness. We employ a convolutional neural network (CNN) to select high-priority targets. The LOW-Z survey has already obtained over 22,000 redshifts of dwarf galaxies (M$_* < 10^9$ M$_\odot$), comparable to the number of dwarf galaxies discovered in SDSS-DR8 and GAMA. As a spare fiber survey, LOW-Z currently receives fiber allocation for just ~50% of its targets. However, we estimate that our selection is highly complete: for galaxies at $z < 0.03$ within our magnitude limits, we achieve better than 95% completeness with ~1% efficiency using catalog-level photometric cuts. We also demonstrate that our CNN selections $z<0.03$ galaxies from the photometric cuts subsample at least ten times more efficiently while maintaining high completeness. The full five-year DESI program will expand the LOW-Z sample, densely mapping the low-redshift Universe, providing an unprecedented sample of dwarf galaxies, and providing critical information about how to pursue effective and efficient low-redshift surveys.

astro-ph.GA↗

arXiv2023

A variable active galactic nucleus at $z=2.06$ triply-imaged by the galaxy cluster MACS J0035.4-2015

We report the discovery of a triply imaged active galactic nucleus (AGN), lensed by the galaxy cluster MACS J0035.4-2015 ($z_{\mathrm{d}}=0.352$). The object is detected in Hubble Space Telescope imaging taken for the RELICS program. It appears to have a quasi-stellar nucleus consistent with a point-source, with a de-magnified radius of $r_e\lesssim100$ pc. The object is spectroscopically confirmed to be an AGN at $z_{\mathrm{spec}}=2.063\pm0.005$ showing broad rest-frame UV emission lines, and is detected in both X-ray observations with Chandra and in ALCS ALMA band 6 (1.2 mm) imaging. It has a relatively faint rest-frame UV luminosity for a quasar-like object, $M_{\mathrm{UV},1450}=-19.7\pm0.2$. The object adds to just a few quasars or other X-ray sources known to be multiply lensed by a galaxy cluster. Some diffuse emission from the host galaxy is faintly seen around the nucleus and there is a faint object nearby sharing the same multiple-imaging symmetry and geometric redshift, possibly an interacting galaxy or a star-forming knot in the host. We present an accompanying lens model, calculate the magnifications and time delays, and infer physical properties for the source. We find the rest-frame UV continuum and emission lines to be dominated by the AGN, and the optical emission to be dominated by the host galaxy of modest stellar mass $M_{\star}\simeq10^{9.2} \mathrm{M}_{\odot}$. We also observe some variation in the AGN emission with time, which may suggest that the AGN used to be more active. This object adds a low-redshift counterpart to several relatively faint AGN recently uncovered at high redshifts with HST and JWST.

astro-ph.GA↗

arXiv2023

Identification of galaxy shreds in large photometric catalogs using Convolutional Neural Networks

Contamination from galaxy fragments, identified as sources, is a major issue in large photometric galaxy catalogs. In this paper, we prove that this problem can be easily addressed with computer vision techniques. We use image cutouts to train a convolutional neural network (CNN) to identify catalogued sources that are in reality just star formation regions and/or shreds of larger galaxies. The CNN reaches an accuracy ~98% on our testing datasets. We apply this CNN to galaxy catalogs from three amongst the largest surveys available today: the Sloan Digital Sky Survey (SDSS), the DESI Legacy Imaging Surveys and the Panoramic Survey Telescope and Rapid Response System Survey (Pan-STARSS). We find that, even when strict selection criteria are used, all catalogs still show a ~5% level of contamination from galaxy shreds. Our CNN gives a simple yet effective solution to clean galaxy catalogs from these contaminants.

astro-ph.GA↗

arXiv2022

Identifying AGN host galaxies with convolutional neural networks

Active galactic nuclei (AGN) are supermassive black holes with luminous accretion disks found in some galaxies, and are thought to play an important role in galaxy evolution. However, traditional optical spectroscopy for identifying AGN requires time-intensive observations. We train a convolutional neural network (CNN) to distinguish AGN host galaxies from non-active galaxies using a sample of 210,000 Sloan Digital Sky Survey galaxies. We evaluate the CNN on 33,000 galaxies that are spectrally classified as composites, and find correlations between galaxy appearances and their CNN classifications, which hint at evolutionary processes that affect both galaxy morphology and AGN activity. With the advent of the Vera C. Rubin Observatory, Nancy Grace Roman Space Telescope, and other wide-field imaging telescopes, deep learning methods will be instrumental for quickly and reliably shortlisting AGN samples for future analyses.

astro-ph.GA↗