Research

Modern astronomical surveys collect complex, high-dimensional datasets across billions of sources, straining traditional statistical methods. Hierarchical Bayesian models capture physical structure, measurement errors, and selection effects, yet exact posterior computation rarely scales to catalog-wide analyses. Conversely, deep learning models scale efficiently to massive survey volumes while often lacking calibrated uncertainty quantification. Research in the ART bridges these approaches through new Markov chain Monte Carlo and nested sampling algorithms, variational and simulation-based inference, conformal prediction, and forward-modelling frameworks where selection functions and observational uncertainties are treated alongside the underlying astrophysical model.

Wide-field photometric, spectroscopic, and time-domain surveys have revealed the rich complexity of stellar interiors, chemical mixing, magnetic activity, and binary evolutionary pathways. Together with gravitational-wave observations from detectors such as LIGO that probe compact stellar remnants (white dwarfs, neutron stars, and black holes), these datasets have transformed our view of how stars live and die. Members of the ART apply hierarchical Bayesian models, hidden Markov models, and machine learning across large survey catalogs to characterize magnetically active stars, gyrochronology and M-dwarf ages, ancient metal-poor stars, binary-stripped helium stars, and asteroseismic populations.

Dark matter halos govern the gravitational potential in which galaxies form and evolve, yet the particle nature and small-scale distribution of dark matter remain open questions in fundamental physics. Members of the ART develop statistical frameworks to measure dark matter halo masses and density profiles across the Milky Way, dwarf satellites, elliptical galaxies, and ultra-diffuse galaxies, testing structure-formation models against observations from Gaia, JWST, DESI, and the Vera C. Rubin Observatory (LSST). On cosmological scales, the group cross-correlates multi-tracer galaxy and supernova surveys to constrain cosmic structure growth and evaluate cosmological tensions with rigorous uncertainty quantification.

Machine learning is now central to astronomical discovery, from classifying millions of galaxy images and spectra to identifying rare gravitational lenses and spectral anomalies. Scientific inference, however, requires more than point predictions. Models used for physical measurement must incorporate domain symmetries, account for observational noise and selection biases, and provide calibrated uncertainties. Members of the ART design physics-informed neural architectures, multimodal foundation models, and interpretable machine learning methods that enable astronomers to search, compare, and analyze petabyte-scale survey archives with statistical rigour.

The Milky Way and its neighbours in the Local Group are the only galaxies where we can resolve billions of individual stars while simultaneously mapping large-scale galactic structure. Measuring stellar positions, three-dimensional velocities, ages, and chemical abundances links the microphysics of stellar populations directly to the assembly history of the Galaxy. Interstellar gas and dust also thread the Galactic disk, attenuating and reddening starlight while supplying the raw material for subsequent star formation. Members of the ART combine chemodynamical modelling, astrometric and spectroscopic inference, and three-dimensional interstellar dust mapping to reconstruct the formation history of the Milky Way and place the Local Group in its broader cosmological context.

Galaxy formation and evolution are driven by the interplay of gas accretion, star formation, stellar feedback, mergers, and supermassive black holes across cosmic time. Because each galaxy is observed at a single epoch, disentangling the physical drivers of galaxy evolution requires large statistical samples and careful modelling of confounding variables and selection effects. Members of the ART apply hierarchical Bayesian modelling, spatial point processes, and spectral energy distribution fitting to observations ranging from nearby globular cluster systems and ultra-diffuse galaxies to ultra-deep JWST spectroscopy of massive galaxies in the early Universe.

Time-domain surveys monitor the dynamic sky across wavelengths, capturing supernovae, stellar flares, and millisecond-duration fast radio bursts (FRBs) from extragalactic sources. Distinguishing repeating from non-repeating FRB populations, characterizing host-galaxy environments, and inferring progenitor physics from brief, censored time series pose core statistical challenges. Members of the ART collaborate within the Canadian Hydrogen Intensity Mapping Experiment (CHIME) FRB project, Rubin Observatory supernova cosmology programs, and high-resolution spectroscopic and radio technosignature searches to model rare transient phenomena and their selection functions.

Stars form when turbulent molecular clouds collapse under self-gravity, while radiation, winds, and supernovae from young stars inject energy and momentum back into the surrounding interstellar medium. Because these processes couple sub-parsec cloud cores to kiloparsec-scale galactic structure, connecting theoretical simulations with observations requires statistical methods that operate across spatial and temporal scales. Members of the ART combine three-dimensional dust mapping, kinematic reconstructions of young stellar associations, and machine learning analyses of magnetohydrodynamic simulations to study how molecular clouds assemble, form stars, and disperse.