Seminars

Mario Pasquato 

(IASF - Milano)

7 November 2024, 11:30, Aula Jappelli 

Causal discovery for astronomy

Astrophysics is about making physical models that predict the relations between quantities we measure in the skies. Models have an implicit notion of causation, e.g. AGN feedback affects star formation.

Data-driven approaches such as machine learning, on the other hand, usually do not have such causal relationships baked in: in a way they are 'just curve fitting'.

The field of causal discovery solves this thanks to algorithms capable of detecting causal relations from data, without relying on physical models. 

Here I recount the first application of these methods to astronomical data, which resulted in solving the chicken-and-egg conundrum of SMBH-galaxy coevolution.

In a recent paper we find that SMBH mass causes galaxy properties in spiral galaxies, but galaxy properties cause SMBH mass in elliptical galaxies.

This is in line with our physical expectations, but it has been obtained in a purely data driven way. I will not omit to cover the shortcomings and limitations of these methods, on which I hope to instigate a stimulating discussion with the audience.

Join Zoom Meeting

https://unipd.zoom.us/j/85273198731?pwd=V1ZiNE9ZWUN4ZTdIUzJsTFo0Z3U2Zz09

Meeting ID: 852 7319 8731

Passcode: 208186

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Next scheduled seminars:

November 19th Sylvain Doutè (Institut de Planétologie et d'Astrophysique de Grenoble, CNRS-UGA, Grenoble.)

and many more! 

HIGH-PROFILE Colloquia: 

December 12th - Thomas BENSBY (University of Lunds) 

January 16th - Adriano FONTANA (INAF) 

February 13th - Claudia CICONE (University of Oslo) 

March 13th - Sara ELLISON (University of Victoria) 

CONTAMINATION Seminars: 

February 6th - Samuele NEGRO (Neurobiologia - UNIPD) 

March 27th - Giacomo DE ANGELIS (INFN)