Letters of Intent received in 2019

LoI 2021-2094
Machine Learning in Astronomy: Possibilities and Pitfalls

Date: 23 August 2021 to 26 August 2021
Location: 2021 GA, Korea, Rep of
Contact: Ashish Mahabal (aam@astro.caltech.edu)
Coordinating division: Division B Facilities, Technologies and Data Science
Other divisions: Division B Facilities, Technologies and Data Science
Chair of SOC: Ashish Mahabal (Caltech)
Chair of LOC: Arman (Shafieloo)



Machine Learning
Deep Learning
Big data
Sky surveys
Virtual observatory




Last few years have seen tremendous growth of machine learning applications to large datasets from various astronomical surveys and even smaller datasets. The traditional machine learning algorithms (e.g. random forests) have seen a healthy growth in their application. What has really taken off is the application of deep learning (mostly Convolutional Neural Networks, but also Recurrent Neural Networks, Generative Adversarial Networks and so on). While most applications are well founded, at least a few are blind applications without full thought having been given to the applicability. The inherent black boxiness of deep learning does not help.

We propose a symposium on the dissemination and demystification of deep learning techniques. There will be tutorials, complete examples, advanced applications, as well sessions like 'bring your own problem', thus catering to novices as well as seasoned practitioners. In particular we hope to encourage fusion of large datasets to enable exploration of combined parameter spaces that have so far remained unexplored. Using such diverse types as images, time series, and spectra, one of our aims is to establish a list of reproducible "best-practices" solutions for astronomy data.

The symposium will also have sessions on more traditional machine learning, especially elucidating areas where deep learning is an overkill (and also sometimes ill-advised). We will supplement this with other astroinformatics sessions including those on methodology transfer to and from other science (e.g. Earth science, medical sciences etc.), visualization, simulations, astrostatistics, archives, pipelines and so on.

The proposed symposium will thus be a follow-up of IAU 325 held at Sorrento, Italy in 2016 where practitioners of computer science, information science, astronomy, and advanced statistics had descended. Of late the US has seen a marked increase in the following of this field. An IAU symposium will help take it to a world wide audience as well as becoming more mainstream. We will work closely with similar groups in other organizations, e.g the American Astronomical Society (AAS) Working Group (of which the proposer is the chair), the International AstroInformatics Association (IAIA), the International Astrostatistics Association (IAA) etc.

We expect 200-300 attendees including computer scientists. We have put together an expert and diverse SOC with experience in organizing such meetings. The IAU's Commission B3 on Astrostatistics & Astroinformatics has initiated the organization of this symposium, and will continue to provide support to the S.O.C. with publicity, liaison with allied interest groups, and other organizational matters. The Proposer of this Letter of Intent is the Vise president of the IAU/B3C. The SOC has identified possible sources of funding to supplement the contribution from the IAU from industry.

The symposium will provide an environment for the exchange of ideas, methods, software, and technical capabilities, seeking to establish enduring associations between the diverse researchers.

The symposium will be in Buson, S Korea, at the site of the GA. Proposed dates are 23-26 Aug 2021

The Symposium lasts 3.5 days. Some sessions will have tutorials, traditional invited review talks, selected contributed talks, and panel discussions. There will also be a hackathon to hack on new and collaborative ideas, an an unconference to generate such ideas before the hackathon. Contributed poster papers will be available throughout the meeting and will include a juried competition with prize.