ScienceGuardians

ScienceGuardians

Did You Know?

ScienceGuardians protects the academic community

Computational astrophysics, data science and AI/ML in astronomy: A perspective from Indian community

Authors: Prateek Sharma,Bhargav Vaidya,Yogesh Wadadekar,Jasjeet Bagla,Piyali Chatterjee,Shravan Hanasoge,Prayush Kumar,Dipanjan Mukherjee,Ninan Sajeeth Philip,Nishant Singh
Journal: Journal of Astrophysics and Astronomy
Publisher: Springer Science and Business Media LLC
Publish date: 2025-5-17
ISSN: 0973-7758 DOI: 10.1007/s12036-025-10049-9
View on Publisher's Website
Up
0
Down
::

You say your stats on HPC/AI usage in Indian astronomy come from 29 respondents.
That’s a tiny sample size for a national community.
How can you generalize trends for the entire Indian A&A field from fewer than 30 people?
Did you check for sampling bias — e.g., mostly from a few institutes, or only those already using HPC?

You claim that access to advanced HPC leads to higher citation impact, citing Wang et al. 2018.
But that study was about XSEDE in the US.
Do you have any India-specific data to back this up for NSM or other national clusters?
Or are you just assuming the US model applies directly here?

You mention that global astrophysics codes are moving to GPU versions, and Indian researchers use them.
But in Section 3.1, most work listed seems CPU-based.
How many of the “community codes” you mention (PLUTO, Athena, etc.) are actually running efficiently on Indian GPU clusters?
Or is this a future wish, not current reality?

All Replies

Viewing 1 replies (of 1 total)

2 weeks, 2 days ago

 The survey reported in Figures 6 and 7 is based on only 29 respondents. How can the authors claim this represents the “Indian astronomy and astrophysics community” (a community of hundreds of active researchers)? With such a minuscule sample size, no statistical significance or community-wide generalizability can be asserted. Why was the sample size not reported in the main text, and why are no error bars or confidence intervals provided for the percentages in Figure 6 (e.g., 48.3% ongoing AI/ML work)? This appears to be a severe case of overgeneralization from non-representative data.

Viewing 1 replies (of 1 total)

  • You must be logged in to reply to this topic.