BRENDON J. BREWER: “Bayesian inference maximizes the possibility of finding the answer to certain question with a minimum of data”

Brendon J. Brewer, researcher at the University of Auckland (New Zealand), is lecturing at the XXVI Winter School of the Astrophysics Institute of the Canaries. Credit: Miguel Briganti, SMM (IAC).
Advertised on

By ADELINA PASTOR

“The major problem in astrophysics is to try to understand the physical processes which underlie the data we obtain”

“Bayesian inference is a tool to help us puzzle out what, and how much, the Universe is telling us.”

 “The mass of the central black hole in the galaxy Arp 151 is between 3 and 6 million solar masses with degree of certainty of 75%.”

Brendon J. Brewer enjoys teaching. For this researcher at the University of Auckland (New Zealand) the most gratifying moment of his work is when his students surprise him with their ideas. Maybe those attending the XXVI Winter School of the Canary Astrophysics Institute (IAC) will manage to do this. Although he is a specialist in Bayesian inference, he does not see his work as different from that of other astrophysicists... “except that I don´t have to stay up all night at the telescope” he adds, with a twinkle in his eye. His work uses statistics to decipher the Universe.

Question. Your work places you between data mining, statistics and astrophysics, as you search among all the information available in the most efficient way possible. Do we have such a lot of data?

Answer: There is a great deal that we don´t know yet, but we have a surprising quantity of information about the universe. You should bear in mind that every time we obtain an image of the sky we take in a large amount of information. Maybe this is not showing us anything relevant, but each image contains around a million pixels, which is indeed a lot of information. Part of the challenge is to get a computer to process all these data and be able to filter out only those which are interesting. At least this is part of the process of data mining, although that´s not my specialty.

Q: What is your line of investigation, then? 

A: The main problem in astrophysics is to try to understand the physical process which underlies the data that we take. I work in a method which lets me make models of what we think is happening in the universe, so that they can be used to simulate the type of data which we would obtain if those models were correct, and to compare these with those which we actually obtain, so as to understand, for example, processes which occur in a distant galaxy which we cannot observe very clearly.

Q: Talking of galaxies, part of your work is the study of active galaxies

A: Active galaxies, such as quasars, normally have a bright point in the centre. We can measure its brightness and study its spectrum, and when we do this we notice that its characteristics vary with time. Many researchers think that this is due to the variation in time of the quantity of matter which falls into the central supermassive black hole from its surroundings. But I should point out a problem in the study of quasars: although they are in reality quite large, in the images they appear as single points, too small to know what is going on in them. To get round this problem we use indirect methods, such as reverberation mapping, in which we monitor the changes in different parts of the quasar, which are similar but occur at different instants due to the time it takes for light to travel from one part of the system to another. What we do in my research is to make a model of the possible structure of this active galactic nucleus and compare it with the available data, so that we can rule out improbable solutions, and retain only those which lie within the range of high probability.

The problem with studying the universe, in general, is that because we are basing our work on indirect observations usually the data at our disposal are not sufficiently good to tell us with certainty if one idea for a model is better than another. We need to find a way in which the data themselves can show us with what degree of certainty we have to take a given theory as opposed to another, and for that we use practical methods of Bayesian statistics, the study of uncertainty.

Q: The way you put it, it would seem that your work is aimed more towards finding a way to look at the universe, rather than towards the study of the universe per se

A: Yes, in a way. Astrophysicists don’t spend some much time in just observing the Universe, rather they reflect on what they have observed, what they would like to observe, and how to make sense of it all. We spend much more time thinking about the observations, and analyzing them that in the process of observing. If you have an idea about what is happening in a specific star, you need to think about what types of observations would give you answers to your questions “ if I observe the star in this way, and I am right, I will see X, but if I am wrong I will see Y”. This is the real challenge.

Q: Is Bayesian inference the tool to overcome this challenge?

A: Bayesian inference is the tool for what comes next. If I have an idea about what is happening in a star, and obtain data, Bayesian inference helps me to calculate the probability that my original idea was right or wrong. Using the method maximizes the possibility of finding an answer to certain questions with a minimum of data. The key is that as well as calculating the probability that a certain idea is correct, Bayesian inference allows us to update the probability if we obtain new information. This is important because usually observing at the telescope does not give us a direct answer, but yields data which support one, or another, hypothesis.

Q: That is to say that it is a tool for deciphering what the Universe is telling us

A: What it is telling us, and how much it is telling us because maybe it is not telling us very much at a given moment. For example when it´s cloudy most of the information is useless. An analogy which I like particularly, is to compare Bayesian inference with a trial in which you don’t know whether the accused is guilty or not, but as you go on building up evidence you get an increasingly clear idea of his (or her) guilt or innocence. Bayesian inference is a process similar, and we can apply it to any question about the Universe.

Q: And what is the last question you have asked of the universe?

A: The mass of the central black hole of the galaxy Arp 151. The answer is between 3 and 6 million times the mass of the sun, with 75% probability.

Organizing Committee: Andrés Asensio Ramos, Íñigo Arregui, Antonio Aparicio y Rafael Rebolo.

Secretary: Lourdes González.

Contacts: Andrés Asensio Ramos (IAC): aasensio [at] iac.es y 922605238 Íñigo Arregui (IAC): iarregui [at] iac.es y 922605465

Press: Carmen del Puerto: prensa [at] iac.es y 922605208

Previous press release: http://www.iac.es/divulgacion.php?op1=16&id=897

Programme of the Winter School: http://www.iac.es/winterschool/2014/pages/about-the-school/timetable.php

Further information: http://www.iac.es/winterschool/2014/

News type