NASA FINESST Fellow
PhD Student, University of Delaware
I study astrophysical transients and their diversity using data-driven methods. Some of my research topics are on probabilistic modelling of light curves, discovery of unusual phenomena, and physically interpretable representations of photometric and spectroscopic data.
Outside astronomy, I enjoy watching and playing football (soccer), bicycling, and coding for fun (e.g. writing a script to scrape postdoc ads and send me alerts when it contains keywords of my interest! :p )
My PhD work currently follows three connected directions in time-domain astrophysics.
Probabilistic data-driven models based on neural processes for sparse, irregular multiband observations for Rubin LSST, ZTF, TESS and DECAM.
Anomaly detection to find unusual and out-of-distribution objects in large time-domain surveys like Rubin LSST.
Data-driven representations of supernova spectra to separate supernova evolution and diversity from noise, and map learned structures back to supernova physics.
For the full publication list, see Google Scholar or my ADS library.
Attentive Neural Processes learn a population of transient light curves and provide fast, probabilistic reconstruction across multiple bands—designed for the scale and cadence of Rubin-era astronomy.
DiMMAD combines multiple distance metrics from DistClassiPy to calculate an anomaly score which helps find objects out-of-distribution - sources most likely to represent undiscovered phenomena.
DistClassiPy explores how different measures of distance change classification in astronomical feature spaces, matching standard approaches while remaining lightweight and interpretable.
Using Gaia DR3 proper motions and radial velocities, we tested SDSS-derived models of the Milky Way disk and halo stellar kinematics against a much larger and more precise astrometric dataset.
We used multi-modal neural networks to distinguish compact galaxies, stars, and quasars in SDSS imaging.
Using the simulated LSST PLAsTiCC dataset, we built an ensemble of gated recurrent units and dense networks to classify irregular astronomical light curves across multiple transient and variable classes.
If you have any questions about my research, wish to discuss potential collaborations, or have spotted a bug in my code, write to me at chaini@udel.edu.
