Skip to navigation | Skip to main content | Skip to footer
Menu
Search the University of Manchester siteSearch Menu StaffNet
Search type

Turing-Manchester Seminar Series (RESCHEDULED)

18 Mar 2022

Now taking place on 29 March

*Please note: this event was originally scheduled to take place on 1 March but has been rescheduled due to the UCU strike action.*

The Digital Futures team are hosting a series of online talks by our new Turing Fellows for 2021-22. For the final event in this series, new Turing Fellow Christopher Conselice is joined by Turing AI Fellow Anna Scaife.

Christopher Conselice, Professor of Extragalactic Astronomy

Machine learning applications in astrophysics: Large surveys and galaxy properties/evolution Recently the application of machine learning, especially deep learning, to astronomy and astrophysics has become very popular. Open use of large data sets in astronomy makes it ideal for applications of machine learning, yet many of the tools, procedures, and adaptability are still in their early stages. Christopher will present some of their work on these topics including using supervised, unsupervised, and regression analyses for determining galaxy properties and evolution. He will also discuss upcoming astronomy missions which will contain hundreds of millions of galaxies, stars, quasars, defects and other detected objects that can only be classified and studied using machine learning techniques that are in the process of being developed.

Anna Scaife, Professor of Radio Astronomy at Jodrell Bank Centre for Astrophysics

AI in the SKA Era: Challenges for Bayesian Neural Networks in Radio Galaxy Classification.

The expected volume of data from the new generation of scientific facilities such as the Square Kilometre Array (SKA) radio telescope has motivated the expanded use of semi-automatic and automatic machine learning algorithms for scientific discovery in astronomy. In this field, the robust and systematic use of machine learning faces a number of specific challenges including a paucity of labelled data for training (paradoxically, although we have too much data, we don't have enough), a clear understanding of the effect of biases introduced due to observational and intrinsic astrophysical selection effects in the training data and motivating a quantitative statistical representation of outcomes from decisive AI applications. In this seminar Anna will talk specifically about the challenge of recovering well-calibrated uncertainties from Bayesian neural networks when classifying radio galaxies, a canonical example of a radio astronomy AI application. Anna will discuss how both model and  likelihood misspecification can affect this calibration, how these effects potentially contribute to the cold posterior effect seen when building models using real astronomical data and what steps can be taken to address these problems.