Seminars & Events
Data Analytics Seminar Series
Mapping the Diversity of Galaxy Spectra with Deep Unsupervised Machine Learning
Dr. Hossen Teimoorinia (NRC Herzberg / Canadian Astronomy Data Centre)
Date: Thursday 20 October 2022
Time: 3:00 - 4:00 PM (ADT)
Venue: Zoom
Modern spectroscopic surveys of galaxies such as MaNGA consist of millions of diverse spectra covering different regions of thousands of galaxies. We propose and implement a deep unsupervised machine learning method to summarize the entire diversity of MaNGA spectra onto a 15x15 map (DESOM-1), where neighbouring points on the map represent similar spectra. We demonstrate our method as an alternative to conventional full spectral fitting for deriving physical quantities, as well as their full probability distributions, much more efficiently than traditional resource-intensive Bayesian methods. Since spectra are grouped by similarity, the distribution of spectra onto the map for a single galaxy, i.e, its "fingerprint", reveals the presence of distinct stellar populations within the galaxy indicating smoother or episodic star-formation histories. We further map the diversity of galaxy fingerprints onto a second map (DESOM-2). Using galaxy images and independent measures of galaxy morphology, we confirm that galaxies with similar fingerprints have similar morphologies and inclination angles. Since morphological information was not used in the mapping algorithm, relating galaxy morphology to the star-formation histories encoded in the fingerprints is one example of how the DESOM maps can be used to make scientific inferences.
****************************
Galaxy evolution with machine learning: the perilous road from simulations to data
Dr. Viviana Acquaviva (CUNY NYC College of Technology)
Date: Thursday 15 September 2022
Time: 3;00 - 4:00 PM (ADT)
Venue: Online via Zoom To obtain the Zoom link, please contact Shannon Rhode at s.rhode@smu.ca.
Obtaining accurate estimates of galaxy properties, from stellar mass to star formation history to chemical enrichment history, is crucial to solving open problems in galaxy formation and evolution. Traditionally, this has been done by using spectral energy distribution fitting methods on spectroscopic or photometric data; more recently, supervised machine learning methods have also been used, with promising results. One potential roadblock for the ML-based methodology is that identifying a training set for this problem is very hard, because the ground truth is not readily available - we have no way of knowing the true history of how galaxies have formed and evolved. As an alternative, we can train machine learning models on state-of-the-art cosmological simulations. In this talk, I will discuss two general aspects of this problem: one, how we can estimate the performance of such models when they are applied to real data; and two, whether an improved feature representation for spectroscopic and photometric galaxy data can help us increase generalizability and robustness.
****************************
Grism spectroscopy for extragalactic astronomy, a how-to
Dr. Gaël Noirot & Dr. Vicente Estrada-Carpenter (Saint Mary’s University)
Date: Thursday 10 March 2022
Time: 3:00 - 4:00 PM (AST)
Venue: Online via Zoom To obtain the Zoom link, please contact Shannon Rhode at s.rhode@smu.ca
Grism spectroscopy is a very efficient way to obtain up to thousands of galaxy spectra in a single exposure. In this seminar, we will briefly review this observing mode and present reduction and analysis techniques of grism data through practical jupyter notebooks. We will present an introduction to using the grism redshift and line analysis software Grizli (https://grizli.readthedocs.io/en/latest/) as well as examples of spectral energy distribution fitting using both grizli and the optimized sampling algorithm dynesty (https://dynesty.readthedocs.io/en/latest/).
****************************
An Introduction to (Dynamic) Nested Sampling
Dr. Joshua Speagle (University of Toronto)
Date: Thursday 13 January 2022
Time: 3:00 - 4:00 PM (AST)
Venue: Online via Zoom To obtain the Zoom link, please contact Shannon Rhode at s.rhode@smu.ca
I will present a brief introduction to Nested Sampling, a complementary framework to Markov Chain Monte Carlo approaches that is designed to estimate marginal likelihoods (i.e. Bayesian evidences) and posterior distributions. This will include some discussion on the philosophical distinctions and motivations of Nested Sampling, a few ways of understanding why/how it works, some of its pros and cons, and more recent extensions such as Dynamic Nested Sampling. I will also highlight how this can work in practice using the public Python package dynesty.
****************************
Teaching a machine to learn to extract stellar properties from sky surveys
Dr. Sébastien Fabbro (NRC Herzberg – Canadian Astronomy Data Centre)
Date: Thursday 18 November 2021
Time: 2:30 - 3:30 PM (AST)
Venue: Online via Zoom To obtain the Zoom link, please contact Shannon Rhode at s.rhode@smu.ca
Astronomical surveys can enable us to trace the evolutionary history of the local Universe. The ever-growing data size and complexity accompanying the surveys have motivated many of us to revisit traditional methods to extract the stellar properties efficiently. I will present some of the recent developments guided by deep learning that can alleviate bottlenecks in the analysis of galactic archaeology surveys. In particular, I will discuss how we can exploit feedback loops between physical modelling and data-driven approaches to face the interpretability and the uncertainty quantification challenges.
****************************