This week a special mini-symposium will be held in-person at
Marsfield, and on-line, with the three international members of the
ATNF Steering Committee each giving a short presentation.
Anna Scaife (Uni of Manchester)
"Scaling laws and self-supervised learning in AI for astrophysics."
As the current trend in deep-learning tends towards more data, more
compute and more parameters, we ask: how large can astrophysical
deep-learning models get? Using the largest dataset of labelled
astronomy data available, we examine the scaling laws for supervised
problems in astrophysics and conclude that self-supervised learning is
a more promising direction for large-scale deep-learning in astronomy
given the available label volumes; however, we also conclude that
using in-domain (astronomy) data for pre-training results in better
performance for the down-stream tasks from these self-supervised
representations. Finally, I will show how we are building
self-supervised foundation models for radio astronomy and a variety of
the downstream tasks where we have applied them in practice.
Jayaram Chengalur (TIFR)
"The evolution of the HI content of galaxies"
Over cosmic time, galaxies grow by merger, and/or by the accretion of
matter via inflows. As galaxies evolve they also convert their gas
into stars. On a cosmic scale, it is well established that the star
formation peaked about 10 billion years ago and that the average star
formation rate per unit volume has declined sharply since
then. Hydrogen is dominant baryonic component of galaxies, and atomic
hydrogen is also the primary fuel for star formation. Stars form as
the gas cools to become molecular hydrogen, and then cools further and
collapses into stars under self gravity. Understanding the evolution
of the atomic hydrogen content of galaxies is hence key to
understanding the evolution of the star formation rate with cosmic
time. The recent upgrade to the Giant Metrewave Radio Telescope
(GMRT), has allowed significant progress to be made in observing HI in
galaxies at redshifts around 1 using the stacking method. In this
talk, I will discuss some of the results from ongoing atomic hydrogen
surveys of star forming galaxies using the upgraded GMRT.
Scott Ransom (NRAO)
"New Exotic and Massive Pulsars in Globular Clusters"
In the past five years, the number of known globular cluster pulsars,
most of them of the millisecond variety, has blossomed by more than
130 to a total of over 300. The main reasons were the commissioning of
the extremely sensitive FAST and MeerKAT radio telescopes, and
increased computing power applied to the searches themselves. In this
talk I'll mention some of the fascinating exotic systems that have
been found, discuss the importance of long-term timing of these
systems, and hint at the astrophysics and basic physics those systems
will provide.
