Joshua Selfridge | Searcharxiv

arXiv2026

The JWST Emission Line Survey (JELS): Multi-wavelength properties of Paschen line-emitters at Cosmic Noon

Context: Narrowband JWST/NIRCam selection isolates star-forming galaxies in well-defined redshift slices, while ALMA and archival far-IR/submm data reveal obscured growth. We combine these to measure Paschen-line SFRs, infer nebular extinction from EAZY Av, and probe FIR-to-submm emission via weighted-median stacking, to constrain galaxy growth at cosmic noon. Aims: We characterize stellar, nebular, and FIR properties of Paschen-selected galaxies in COSMOS, in two ensembles at z~1.50 (Pa-alpha) and z~2.65 (Pa-beta), quantifying short- vs long-timescale SF and dust attenuation, and constraining obscured SFR and gas masses. Methods: We select robust narrowband-excess samples from the JELS F466N/F470N mosaic (Pa-alpha: 77, Pa-beta: 31 sources), remove AGN via X-ray/radio matching, and assemble multiwavelength photometry (JWST, HST, Spitzer, Herschel, SCUBA-2, AzTEC, ALMA). Photo-z/stellar properties come from EAZY/STARDUST; line SFRs use recent Paschen calibrations, with nebular extinction from EAZY Av. Results: EAZY fits show stellar masses log(Mstar/Msun)~7.4-10.3 and modest SFRs (0.1-27 Msun/yr). Line SFRs are 0.1-22.0 and 0.7-47.9 Msun/yr for Pa-alpha/beta. Most sources show short-term bursts (line SFRs exceed SED SFRs), weakly decreasing with mass. Nebular attenuation correlates with mass and SFR. Stacking gives no >3sigma detections beyond MIRI. The weighted-median SED gives SFR_UV,med = 0.48+-0.11 and 3.92+-0.47 Msun/yr for Pa-alpha/beta. ALMA Band 4 gives the tightest constraints: median 3-sigma limits SFR_IR<53 (Pa-alpha) and <44 Msun/yr (Pa-beta); gas mass <10^11.2 (Pa-alpha) and <10^10.6 Msun (Pa-beta). Conclusions: Paschen-selected samples show widespread short-timescale star formation, but FIR/sub-mm data give only upper limits on obscured SFR/ISM mass. Deeper spectroscopy and ALMA continuum data are needed to constrain obscured SF and gas reservoirs.

astro-ph.GA↗

arXiv2024

Parallelizing non-linear sequential models over the sequence length

Sequential models, such as Recurrent Neural Networks and Neural Ordinary Differential Equations, have long suffered from slow training due to their inherent sequential nature. For many years this bottleneck has persisted, as many thought sequential models could not be parallelized. We challenge this long-held belief with our parallel algorithm that accelerates GPU evaluation of sequential models by up to 3 orders of magnitude faster without compromising output accuracy. The algorithm does not need any special structure in the sequential models' architecture, making it applicable to a wide range of architectures. Using our method, training sequential models can be more than 10 times faster than the common sequential method without any meaningful difference in the training results. Leveraging this accelerated training, we discovered the efficacy of the Gated Recurrent Unit in a long time series classification problem with 17k time samples. By overcoming the training bottleneck, our work serves as the first step to unlock the potential of non-linear sequential models for long sequence problems.

cs.LG↗