Machine learning to maximise the impact of ALMA

Lead Research Organisation: Cardiff University
Department Name: School of Physics and Astronomy

Abstract

The Atacama Large Millimeter/submillimeter Array (ALMA) is the largest scale operating telescope project in the world. This revolutionary facility is producing over a petabyte of data a year, much of which will never be inspected by eye. Machine learning and data based approaches to exploiting this vast data archive are clearly required, For instance developing robust unsupervised algorithms for automated source/line detection, and innovative calibration algorithms based on machine learning.

Publications

10 25 50
publication icon
Dawson J (2019) Using machine learning to study the kinematics of cold gas in galaxies in Monthly Notices of the Royal Astronomical Society

Studentship Projects

Project Reference Relationship Related To Start End Student Name
ST/P006779/1 01/10/2017 30/09/2024
1945968 Studentship ST/P006779/1 01/10/2017 30/09/2021 James James Dawson
 
Title Optimisation and release of KinMS Python tool 
Description I am a leader in the recent boost in developing KinMSpy, a Python tool for astronomers modelling the kinematics of molecular gas in astronomy. Along with a colleague we have improved the speed of this code and provided stable documentation as well as a PyPi package release. 
Type Of Material Improvements to research infrastructure 
Year Produced 2020 
Provided To Others? Yes  
Impact Several requests have begun arriving for using the tool and there are ongoing debugging requests. 
URL https://github.com/TimothyADavis/KinMSpy/tree/dev
 
Title Machine learning model for position angle and circularity predictions of CO gas in galaxies 
Description This model allows users to analyse their galaxy observations from interferometers such as ALMA, VLA, SKA, etc. They can then retrieve information on the level of ordered rotation of the gas in those galaxies and their position angles for further modelling. 
Type Of Material Computer model/algorithm 
Year Produced 2019 
Provided To Others? Yes  
Impact Several conference talks have been accepted on the topic and this work forms the groundwork for my next publication using more advanced machine learning methods. 
URL https://github.com/SpaceMeerkat/CAE
 
Description Using machine learning to study the kinematics of cold gas in galaxies 
Organisation RWTH Aachen University
Country Germany 
Sector Academic/University 
PI Contribution Principle investigator on the project leading to 2+ papers on using machine learning to prepare for the SKA.
Collaborator Contribution PyTorch guidance and paper reviews by Justus Schock.
Impact MNRAS paper accepted and published under the title "Using machine learning to study the kinematics of cold gas in galaxies".
Start Year 2019
 
Title KinMSpy active development 
Description The KinMS (KINematic Molecular Simulation) package can be used to simulate observations of arbitary molecular/atomic cold gas distributions. The routines are written with flexibility in mind, and have been used in various different applications, including investigating the kinematics of molecular gas in early-type galaxies (Davis et al, MNRAS, Volume 429, Issue 1, p.534-555, 2013), and determining supermassive black-hole masses from CO interfermetric observations (Davis et al., Nature, 2013). They are also useful for creating input datacubes for further simulation in e.g. CASA's sim_observe tool. 
Type Of Technology Software 
Year Produced 2020 
Open Source License? Yes  
Impact Several requests to use the tool and regular bug updates. 
URL https://github.com/TimothyADavis/KinMSpy/tree/dev
 
Description Taking part in the ARIEL machine learning challenge 2019 and presenting at ECML PKDD 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Study participants or study members
Results and Impact In 2019 I won ARIEL machine learning challenge. This involved removing sunspot noise from exoplanet transit light-curves and using neural networks to predict their corresponding radii. In 2020 I presented this work at ECML PKDD. Around 50-100 people attended the conference and I am currently engaged in writing a joint paper on the results.
Year(s) Of Engagement Activity 2019
URL https://ariel-datachallenge.azurewebsites.net/ML