arXiv subjects

Sean Moran

Publications and source records attributed to Sean Moran.

At least 37 records · Page 2Linked to original sources

arXiv2023

Code Librarian: A Software Package Recommendation System

The use of packaged libraries can significantly shorten the software development cycle by improving the quality and readability of code. In this paper, we present a recommendation engine called Librarian for open source libraries. A candidate library package is recommended for a given context if: 1) it has been frequently used with the imported libraries in the program; 2) it has similar functionality to the imported libraries in the program; 3) it has similar functionality to the developer's implementation, and 4) it can be used efficiently in the context of the provided code. We apply the state-of-the-art CodeBERT-based model for analysing the context of the source code to deliver relevant library recommendations to users.

cs.SE↗

arXiv2022

Senatus -- A Fast and Accurate Code-to-Code Recommendation Engine

Machine learning on source code (MLOnCode) is a popular research field that has been driven by the availability of large-scale code repositories and the development of powerful probabilistic and deep learning models for mining source code. Code-to-code recommendation is a task in MLOnCode that aims to recommend relevant, diverse and concise code snippets that usefully extend the code currently being written by a developer in their development environment (IDE). Code-to-code recommendation engines hold the promise of increasing developer productivity by reducing context switching from the IDE and increasing code-reuse. Existing code-to-code recommendation engines do not scale gracefully to large codebases, exhibiting a linear growth in query time as the code repository increases in size. In addition, existing code-to-code recommendation engines fail to account for the global statistics of code repositories in the ranking function, such as the distribution of code snippet lengths, leading to sub-optimal retrieval results. We address both of these weaknesses with \emph{Senatus}, a new code-to-code recommendation engine. At the core of Senatus is \emph{De-Skew} LSH a new locality sensitive hashing (LSH) algorithm that indexes the data for fast (sub-linear time) retrieval while also counteracting the skewness in the snippet length distribution using novel abstract syntax tree-based feature scoring and selection algorithms. We evaluate Senatus and find the recommendations to be of higher quality than competing baselines, while achieving faster search. For example on the CodeSearchNet dataset Senatus improves performance by 31.21\% F1 and 147.9\emph{x} faster query time compared to Facebook Aroma. Senatus also outperforms standard MinHash LSH by 29.2\% F1 and 51.02\emph{x} faster query time.

cs.SE↗

arXiv2022

ST-FL: Style Transfer Preprocessing in Federated Learning for COVID-19 Segmentation

Chest Computational Tomography (CT) scans present low cost, speed and objectivity for COVID-19 diagnosis and deep learning methods have shown great promise in assisting the analysis and interpretation of these images. Most hospitals or countries can train their own models using in-house data, however empirical evidence shows that those models perform poorly when tested on new unseen cases, surfacing the need for coordinated global collaboration. Due to privacy regulations, medical data sharing between hospitals and nations is extremely difficult. We propose a GAN-augmented federated learning model, dubbed ST-FL (Style Transfer Federated Learning), for COVID-19 image segmentation. Federated learning (FL) permits a centralised model to be learned in a secure manner from heterogeneous datasets located in disparate private data silos. We demonstrate that the widely varying data quality on FL client nodes leads to a sub-optimal centralised FL model for COVID-19 chest CT image segmentation. ST-FL is a novel FL framework that is robust in the face of highly variable data quality at client nodes. The robustness is achieved by a denoising CycleGAN model at each client of the federation that maps arbitrary quality images into the same target quality, counteracting the severe data variability evident in real-world FL use-cases. Each client is provided with the target style, which is the same for all clients, and trains their own denoiser. Our qualitative and quantitative results suggest that this FL model performs comparably to, and in some cases better than, a model that has centralised access to all the training data.

eess.IV↗

arXiv2021

Transforming gas-rich low-mass disky galaxies into ultra-diffuse galaxies by ram pressure

Faint extended elliptically-shaped ultra-diffuse galaxies and slightly brighter and more compact dwarf elliptical and lenticular stellar systems are common in galaxy clusters. Their poorly constrained evolutionary paths can be studied by identifying young ultra-diffuse galaxy and dwarf elliptical analogs populated with bright, massive stars. Using data mining we identified 11 such low-mass ($2\times10^8<M_*<2\times10^9 M_{\odot}$) galaxies with large half-light radii ($2.0<r_e<5$ kpc) and recently quenched star formation in the Coma and Abell 2147 galaxy clusters. All galaxies happen to have ram-pressure-stripped tails with signs of current or recent star formation. Deep spectroscopic observations revealed rotating stellar discs containing 70-95% dark matter by mass. A large fraction of the disc stars (10-60%) formed in intense star bursts 180-970 Myr ago, probably triggered by ram pressure. Observed global gradients of stellar age corroborate this scenario. Passive evolution in the next 10 Gyr will transform 9 of the 11 galaxies into ultra-diffuse galaxies. If we assume a constant rate of galaxy infall, 44$\pm$16% of the most luminous present-day ultra-diffuse galaxies in Coma must have formed via ram pressure stripping of disky progenitors.

astro-ph.GA↗

arXiv2020

Center for Astrophysics Optical Infrared Science Archive. I. FAST Spectrograph

We announce the public release of 141,531 moderate-dispersion optical spectra of 72,247 objects acquired over the past 25 years with the FAST Spectrograph on the Fred L. Whipple Observatory 1.5-meter Tillinghast telescope. We describe the data acquisition and processing so that scientists can understand the spectra. We highlight some of the largest FAST survey programs, and make recommendations for use. The spectra have been placed in a Virtual Observatory accessible archive and are ready for download.

astro-ph.IM↗

arXiv2020

CURL: Neural Curve Layers for Global Image Enhancement

We present a novel approach to adjust global image properties such as colour, saturation, and luminance using human-interpretable image enhancement curves, inspired by the Photoshop curves tool. Our method, dubbed neural CURve Layers (CURL), is designed as a multi-colour space neural retouching block trained jointly in three different colour spaces (HSV, CIELab, RGB) guided by a novel multi-colour space loss. The curves are fully differentiable and are trained end-to-end for different computer vision problems including photo enhancement (RGB-to-RGB) and as part of the image signal processing pipeline for image formation (RAW-to-RGB). To demonstrate the effectiveness of CURL we combine this global image transformation block with a pixel-level (local) image multi-scale encoder-decoder backbone network. In an extensive experimental evaluation we show that CURL produces state-of-the-art image quality versus recently proposed deep learning approaches in both objective and perceptual metrics, setting new state-of-the-art performance on multiple public datasets. Our code is publicly available at: https://github.com/sjmoran/CURL.

eess.IV↗

arXiv2020

Low Light Video Enhancement using Synthetic Data Produced with an Intermediate Domain Mapping

Advances in low-light video RAW-to-RGB translation are opening up the possibility of fast low-light imaging on commodity devices (e.g. smartphone cameras) without the need for a tripod. However, it is challenging to collect the required paired short-long exposure frames to learn a supervised mapping. Current approaches require a specialised rig or the use of static videos with no subject or object motion, resulting in datasets that are limited in size, diversity, and motion. We address the data collection bottleneck for low-light video RAW-to-RGB by proposing a data synthesis mechanism, dubbed SIDGAN, that can generate abundant dynamic video training pairs. SIDGAN maps videos found 'in the wild' (e.g. internet videos) into a low-light (short, long exposure) domain. By generating dynamic video data synthetically, we enable a recently proposed state-of-the-art RAW-to-RGB model to attain higher image quality (improved colour, reduced artifacts) and improved temporal consistency, compared to the same model trained with only static real video data.

eess.IV↗

arXiv2020

DeepLPF: Deep Local Parametric Filters for Image Enhancement

Digital artists often improve the aesthetic quality of digital photographs through manual retouching. Beyond global adjustments, professional image editing programs provide local adjustment tools operating on specific parts of an image. Options include parametric (graduated, radial filters) and unconstrained brush tools. These highly expressive tools enable a diverse set of local image enhancements. However, their use can be time consuming, and requires artistic capability. State-of-the-art automated image enhancement approaches typically focus on learning pixel-level or global enhancements. The former can be noisy and lack interpretability, while the latter can fail to capture fine-grained adjustments. In this paper, we introduce a novel approach to automatically enhance images using learned spatially local filters of three different types (Elliptical Filter, Graduated Filter, Polynomial Filter). We introduce a deep neural network, dubbed Deep Local Parametric Filters (DeepLPF), which regresses the parameters of these spatially localized filters that are then automatically applied to enhance the image. DeepLPF provides a natural form of model regularization and enables interpretable, intuitive adjustments that lead to visually pleasing results. We report on multiple benchmarks and show that DeepLPF produces state-of-the-art performance on two variants of the MIT-Adobe-5K dataset, often using a fraction of the parameters required for competing methods.

cs.CV↗

arXiv2019

Evolutionary link between ultra-diffuse galaxies and dwarf early-type galaxies

Spectroscopic studies of low-luminosity early-type galaxies are essential to understand their origin and evolution but remain challenging because of low surface brightness levels. We describe an observational campaign with the new high-throughput Binospec spectrograph at the 6.5-m MMT. It targets a representative sample of dwarf elliptical (dE), ultra-diffuse (UDG), and dwarf spheroidal (dSph) galaxies. We outline our data analysis approach that features (i) a full spectrophotometric fitting to derive internal kinematics and star formation histories of galaxies; (ii) two-dimensional light profile decomposition; (iii) Jeans anisotropic modelling to assess their internal dynamics and dark matter content. We present first results for 9 UDGs in the Coma cluster and a nearby dSph galaxy, which suggest that a combination of internal (supernovae feedback) and environmental (ram-pressure stripping, interactions) processes can explain observed properties of UDGs and, therefore, establish an evolutionary link between UDGs, dSph, and dE galaxies.

astro-ph.GA↗

arXiv2019

NODE: Extreme Low Light Raw Image Denoising using a Noise Decomposition Network

Denoising extreme low light images is a challenging task due to the high noise level. When the illumination is low, digital cameras increase the ISO (electronic gain) to amplify the brightness of captured data. However, this in turn amplifies the noise, arising from read, shot, and defective pixel sources. In the raw domain, read and shot noise are effectively modelled using Gaussian and Poisson distributions respectively, whereas defective pixels can be modeled with impulsive noise. In extreme low light imaging, noise removal becomes a critical challenge to produce a high quality, detailed image with low noise. In this paper, we propose a multi-task deep neural network called Noise Decomposition (NODE) that explicitly and separately estimates defective pixel noise, in conjunction with Gaussian and Poisson noise, to denoise an extreme low light image. Our network is purposely designed to work with raw data, for which the noise is more easily modeled before going through non-linear transformations in the image signal processing (ISP) pipeline. Quantitative and qualitative evaluation show the proposed method to be more effective at denoising real raw images than state-of-the-art techniques.

eess.IV↗

arXiv2019

Internal dynamics and stellar content of nine ultra-diffuse galaxies in the Coma cluster prove their evolutionary link with dwarf early-type galaxies

Ultra-diffuse galaxies (UDGs) are spatially extended, low surface brightness stellar systems with regular elliptical-like morphology found in a wide range of environments. Studies of the internal dynamics and dark matter content of UDGs that would elucidate their formation and evolution have been hampered by their low surface brightnesses. Here we present spatially resolved velocity profiles, stellar velocity dispersions, ages and metallicities for 9 UDGs in the Coma cluster. We use intermediate-resolution spectra obtained with Binospec, the MMT's new high-throughput optical spectrograph. We derive dark matter fractions between 50~\%\ and 90~\% within the half-light radius using Jeans dynamical models. Three galaxies exhibit major axis rotation, two others have highly anisotropic stellar orbits, and one shows signs of triaxiality. In the Faber--Jackson and mass--metallicity relations, the 9 UDGs fill the gap between cluster dwarf elliptical (dE) and fainter dwarf spheroidal (dSph) galaxies. Overall, the observed properties of all 9 UDGs can be explained by a combination of internal processes (supernovae feedback) and environmental effects (ram-pressure stripping, interaction with neighbors). These observations suggest that UDGs and dEs are members of the same galaxy population.

astro-ph.GA↗

arXiv2019

Mapping the Stellar Halo with the H3 Spectroscopic Survey

Modern theories of galaxy formation predict that the Galactic stellar halo was hierarchically assembled from the accretion and disruption of smaller systems. This hierarchical assembly is expected to produce a high degree of structure in the combined phase and chemistry space; this structure should provide a relatively direct probe of the accretion history of our Galaxy. Revealing this structure requires precise 3D positions (including distances), 3D velocities, and chemistry for large samples of stars. The Gaia satellite is delivering proper motions and parallaxes for >1 billion stars to G~20. However, radial velocities and metallicities will only be available to G~15, which is insufficient to probe the outer stellar halo (>10 kpc). Moreover, parallaxes will not be precise enough to deliver high-quality distances for stars beyond ~10 kpc. Identifying accreted systems throughout the stellar halo therefore requires a large ground-based spectroscopic survey to complement Gaia. Here we provide an overview of the H3 Stellar Spectroscopic Survey, which will deliver precise stellar parameters and spectrophotometric distances for 200,000 stars to r=18. Spectra are obtained with the Hectochelle instrument at the MMT, which is configured for the H3 Survey to deliver resolution R~23,000 spectra covering the wavelength range 5150A-5300A. The survey is optimized for stellar halo science and therefore focuses on high Galactic latitude fields (|b|>30 deg.), sparsely sampling 15,000 sq. degrees. Targets are selected on the basis of Gaia parallaxes, enabling very efficient selection of bone fide halo stars. The survey began in the Fall of 2017 and has collected 88,000 spectra to-date. All of the data, including the derived stellar parameters, will eventually be made publicly available via the survey website: h3survey.rc.fas.harvard.edu.

astro-ph.GA↗

arXiv2019

Binospec Software System

Binospec is a high-throughput, 370 to 1000 nm, imaging spectrograph that addresses two adjacent 8' by 15' fields of view. Binospec was commissioned in late 2017 at the f/5 focus of the 6.5m MMT and is now available to all MMT observers. Here we describe the Binospec software used for observation planning, instrument control, and data reduction. The software and control systems incorporate a high level of automation to minimize observer workload. Instrument configuration and observation sequencing is implemented using a database-driven approach to maximize observatory efficiency. A web-based interface allows users to define observations, monitor status, and retrieve data products.

astro-ph.IM↗

arXiv2016

HectoMAP and Horizon Run 4: Dense Structures and Voids in the Real and Simulated Universe

HectoMAP is a dense redshift survey of red galaxies covering a 53 $deg^{2}$ strip of the northern sky. HectoMAP is 97\% complete for galaxies with $r<20.5$, $(g-r)>1.0$, and $(r-i)>0.5$. The survey enables tests of the physical properties of large-scale structure at intermediate redshift against cosmological models. We use the Horizon Run 4, one of the densest and largest cosmological simulations based on the standard $Λ$ Cold Dark Matter ($Λ$CDM) model, to compare the physical properties of observed large-scale structures with simulated ones in a volume-limited sample covering 8$\times10^6$ $h^{-3}$ Mpc$^3$ in the redshift range $0.22<z<0.44$. We apply the same criteria to the observations and simulations to identify over- and under-dense large-scale features of the galaxy distribution. The richness and size distributions of observed over-dense structures agree well with the simulated ones. Observations and simulations also agree for the volume and size distributions of under-dense structures, voids. The properties of the largest over-dense structure and the largest void in HectoMAP are well within the distributions for the largest structures drawn from 300 Horizon Run 4 mock surveys. Overall the size, richness and volume distributions of observed large-scale structures in the redshift range $0.22<z<0.44$ are remarkably consistent with predictions of the standard $Λ$CDM model.

astro-ph.CO↗

arXiv2015

Data Reduction Pipeline for the MMT and Magellan Infrared Spectrograph

We describe the new spectroscopic data reduction pipeline for the multi-object MMT/Magellan Infrared Spectrograph. The pipeline is implemented in idl as a stand-alone package and is publicly available in both stable and development versions. We describe novel algorithms for sky subtraction and correction for telluric absorption. We demonstrate that our sky subtraction technique reaches the Poisson limit set by the photon statistics. Our telluric correction uses a hybrid approach by first computing a correction function from an observed stellar spectrum, and then differentially correcting it using a grid of atmosphere transmission models for the target airmass value. The pipeline provides a sufficient level of performance for real time reduction and thus enables data quality control during observations. We reduce an example dataset to demonstrate the high data reduction quality.

astro-ph.IM↗

arXiv2015

A systematic study of the inner rotation curves of galaxies observed as part of the GASS and COLD GASS surveys

We present a systematic analysis of the rotation curves of 187 galaxies with masses greater than 10^10 M_sol, with atomic gas masses from the GALEX Arecibo Sloan Survey (GASS), and with follow-up long-slit spectroscopy from the MMT. Our analysis focuses on stellar rotation curves derived by fitting stellar template spectra to the galaxy spectra binned along the slit. In this way, we are able to obtain accurate rotation velocity measurements for a factor of 2 more galaxies than possible with the Halpha line. Galaxies with high atomic gas mass fractions are the most dark-matter dominated galaxies in our sample and have dark matter halo density profiles that are well fit by Navarro, Frenk & White profiles with an average concentration parameter of 10. The inner slopes and of the rotation curves correlate more strongly with stellar population age than with galaxy mass or structural parameters. At fixed stellar mass, the rotation curves of more actively star-forming galaxies have steeper inner slopes than less actively star-forming galaxies. The ratio between the galaxy specific angular momentum and the total specific angular momentum of its dark matter halo, R_j, correlates strongly with galaxy mass, structure and gas content. Low mass, disk-dominated galaxies with atomic gas mass fractions greater than 20% have median values of R_j of around 1, but massive, bulge-dominated galaxies have R_j=0.2-0.3. We argue that these trends can be understood in a picture where gas inflows triggered by disk instabilities lead to the formation of passive, bulge-dominated galaxies with low specific angular momentum.

astro-ph.GA↗

arXiv2013

A Multi-Wavelength Analysis of NGC 4178: A Bulgeless Galaxy with an AGN

We present {\it Gemini} longslit optical spectroscopy and VLA radio observations of the nuclear region of NGC 4178, a late-type bulgeless disk galaxy recently confirmed to host an AGN through infrared and X-ray observations. Our observations reveal that the dynamical center of the galaxy is coincident with the location of the {\it Chandra} X-ray point source discovered in a previous work, providing further support for the presence of an AGN. While the X-ray and IR observations provide robust evidence for an AGN, the optical spectrum shows no evidence for the AGN, underscoring the need for the penetrative power of mid-IR and X-ray observations in finding buried or weak AGNs in this class of galaxy. Finally, the upper limit to the radio flux, together with our previous X-ray and IR results, is consistent with the scenario in which NGC 4178 harbors a deeply buried AGN accreting at a high rate.

astro-ph.CO↗

arXiv2013

Star Formation and Metallicity Gradients in Semi-analytic Models of Disk Galaxy Formation

We have updated our radially-resolved SAMs of galaxy formation, which track both the atomic and molecular gas phases of the ISM. The models are adapted from those of Guo et al. using similar methodology as in Fu et al. and are run on halo merger trees from the MS and MS II with the following main changes: (1) We adopt a simple star formation law where Σ_SFR \propto Σ_H2 (2) We inject the heavy elements produced by supernovae directly into the halo hot gas, instead of first mixing them with the cold gas in the disk. (3) We include radial gas inflows in disks using a model of the form v_inflow = αr. The models are used to study the radial profiles of star formation rate and gas-phase metallicity in present-day galaxies. The Σ_H2 profiles in L* galaxies place strong constraints on inflow velocities, favouring models where v_inflow~7km/s at a galactocentric radius of 10kpc. Radial gas inflow has little influence on gas-phase and stellar metallicity gradients, which are affected much more strongly by the fraction of metals that are directly injected into the halo gas, rather than mixed with the cold gas. Metals ejected out of the galaxy in early epochs result in late infall of pre-enriched gas and flatter present-day gas-phase metallicity gradients. A prescription in which 80% of the metals are injected into the halo gas results in good fits to the flat observed metallicity gradients in galaxies with stellar masses greater than 10^10 M_sun, as well as the relations between gas-phase metallicity and sSFR in the outer parts of galactic disks. We examine the correlation between gas-phase metallicity gradient and global galaxy properties, finding that it is most strongly correlated with the bulge-to-total ratio of the galaxy. This is because gas is consumed when the bulge forms during galaxy mergers, and the gas-phase metallicity gradient is then set by newly-accreted gas.

astro-ph.CO↗