Background
Following on from Tony's scistaff meeting about Astronomy AI institutes in August 2023, we decided it would be good to set up a regular (~monthly) series of meetings on using AI/ML focused especially on VLASS, which, as it produces a large, uniform set of images and other products makes it a good candidate for this sort of work, but also covering other areas e.g. science ready data products, pipeline QA etc, particularly with an eye towards what will be needed for ngVLA.
Aim
To draft a white paper/VLASS memo on these topics (on a timescale ~ 1 year) and, if we find some particularly compelling use cases, to include an implementation plan, an estimate of the resources needed to carry it out and outreach to possible partners from the NSF's AI institute effort.
Resources
There exists a rich literature of astronomical uses of AI/ML, here are some interesting papers/talks that might be relevant. See also this set of PASP papers on machine learning in astronomy
Image and image artifact classification
CIRADA (Canadian Initiative for Radio Astronomy Data Analysis)'s use of Self-organizing maps for their VLASS catalogs: CIRADA catalog users guide
Felix Stoehr's ADASS talk on ALMA archive development "You might also like these images"
Brian Kent's demo for the 2023 June AAS splinter session, classifying VLASS sources with a neural network
Point source detection: DEEPSOURCE: point source detection using deep learning
Interferometric imaging with machine learning
https://arxiv.org/abs/2311.06349 - Interferometric Image Reconstruction using Closure Invariants and Machine Learning
Deep radio-interferometric imaging with POLISH: DSA-2000 and weak lensing (Connor et al. 2022; ADS)
First AI for Deep Super-resolution Wide-field Imaging in Radio Astronomy: Unveiling Structure in ESO 137-006 (Dabbech et al. 2022; ADS)
3D detection and characterization of ALMA sources through deep learning ADS Github
ESO's BRAIN study (ALMA dev. program) ArXiv
RFI detection and excision
DSC based Dual-Resunet for radio frequency interference identification
U-net based "A Self-learning Neural Network Approach for Radio Frequency Interference Detection and Removal in Radio Astronomy". Here is the Github repo
Other topics
https://www.newyorker.com/magazine/2023/11/20/a-coder-considers-the-waning-days-of-the-craft
Software resources
Scikit learn (data science scripts in python)
Tensorflow (open-source code originally from Google)
PyTorch (open source code originally from Facebook)
Kaggle (community published models and code)
Meetings/conferences
ESOGPT (September 2024)
Rare gems in big data, Tucson, May 20-23 2024