All attendees are expected to show respect and courtesy to other attendees and staff, and to adhere to the NAM Code of Conduct.
Parallel sessions
Sessions
id
date time
2022-03-08 11:49:00
Machine Learning in Modern Astronomy: Learning and Interpreting the Data Driven Universe
Techniques2
Machine learning (ML) plays a crucial role in modern astronomy and astrophysics, powering robust analyses and visualisations of large/complex datasets and minimising the human effort required to keep pace with the significant streams of data being generated by current-generation projects. These techniques can also be used to accelerate complex numerical simulations, saving significant amounts of compute time and enabling faster analyses. With ~5% of all astronomy papers published in 2021 referencing machine learning (ML) and deep learning (DL) and this figure increasing rapidly each year, it is clear that these techniques are becoming mainstream -- with both supervised, unsupervised, and reinforcement learning techniques finding broad utility across a range of different sub-fields. Of increasing importance as we develop more complex astrophysical models are robust ML resistant to outliers, explainable/interpretable ML, and uncertainty quantification. These areas of research have traditionally proven challenging, yet are crucial for deployment of ML/DL models in large astronomical datasets, and are essential for the responsible use of these models, both in astrophysics and further afield. With the above in mind, this NAM session will discuss the latest innovations in ML/DL from across astronomy and solar physics. The session is intended to encourage valuable cross-pollination of ideas between fields, and provide crucial networking opportunities for early-career researchers that have been lacking due to the current pandemic. We will showcase recent and exciting research from across ML in astronomy and solar physics, with the hope of encouraging fruitful discussions among session participants and highlighting novel techniques to apply to their own research. To promote open-source practices in science, we also encourage speakers to make available short tutorial notebooks demonstrating key techniques from their talk as a takeaway from the session.
Tom Killestein, David Armstrong, Eliot Ayache, Joe Lyman, Azib Norazman, Catarina Sampaio Alves
Wed. 16:30-18:00
16.30-16.35 Thomas Killestein: Introduction
16.35-16.55 Emily Hunt: The power (and caveats) of clustering algorithms applied to Gaia data
16.55-17.10 Timothy A Davis: Self-supervised, physics-aware, Bayesian neural networks for modelling galaxy emission-line kinematics
17.10-17.25 Matthew Mould: Data-driven inference of gravitational-wave populations with machine learning models
17.25-17.40 Alessio Spurio Mancini: COSMOPOWER: Deep Learning – accelerated cosmological inference from next-generation surveys
17.40-17.50 Joshua William Wilde: Where & Why: Interpreting Convolutional Neural Networks Trained to Identify Strong Gravitational Lenses
17.50-18.00 Discussion [panel]