Cosmology and galaxy astrophysics with simulations and machine learning 2024
-
Current and upcoming experiments such as DES, DESI, LSST, Euclid, Roman, and SKA will provide an unprecedented volume of data to constrain the value of the fundamental cosmological parameters and improve our understanding of galaxy evolution across cosmic time. However, maximizing the scientific output of extragalactic surveys requires overcoming substantial challenges: (1) the optimal summary statistic for extracting the maximum amount of cosmological information is unknown for the non-Gaussian fields observed in most surveys, (2) a large amount of cosmological information is out of reach on scales that are significantly affected by galaxy formation physics, and (3) uncertainties in astrophysical processes such as the impact of feedback from massive stars and supermassive black holes, which limit the predictive power of galaxy formation models, are represented by a large number of parameters in various sub-grid models that are not well characterized and understood.
The emergence of powerful machine learning methods provides new approaches to extracting information from data. Trained on large sets of cosmological N-body simulations such as AbacusSummit, Aemulus, DarkQuest, and Quijote, neural networks can search through all possible summary statistics to extract cosmological information at the field level and provide tighter constraints on the value of the cosmological parameters compared to traditional inference from power spectra. Meanwhile, large-volume cosmological hydrodynamic simulations such as IllustrisTNG, EAGLE, ASTRID, HorizonAGN, SIMBA, and FLAMINGO provide the framework to model explicitly the impact of baryonic physics on galaxy formation and large-scale structure, enabling machine learning applications to improve galaxy-halo models and emulate hydrodynamic effects on top of N-body simulations.
Recently, the Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) project has taken a step further by providing thousands of cosmological hydrodynamic simulations with different implementations of baryonic physics and systematic variations of uncertain feedback parameters to train machine learning models that can (1) maximize the extraction of cosmological information while marginalizing over uncertainties in galaxy formation physics, (2) emulate baryonic properties such as the galaxy-halo connection and the large-scale distribution of matter conditioned on astrophysical parameters, and (3) explore the vast space of cosmological and astrophysical parameters and their non-linear interactions governing galaxy evolution.
Following numerous recent results and major public data releases in this field, the goal of this workshop is to bring a diverse community together to (1) review the current state of the use of cosmological simulations and machine learning to address major challenges in galaxy evolution and cosmology, (2) discuss future plans to support upcoming surveys, and (3) foster the exchange of ideas and new collaborations. This workshop will include a mix of invited presentations, contributed talks, and ample discussion time, along with breakout sessions for more in-depth collaborations. The workshop is intended for anyone interested in these topics, whether you are already directly involved in a major collaboration in this field, intend to use public data released as part of projects such as those mentioned above, or are considering starting new activities in this growing field.
-
- Registration and abstract submission for the workshop closed on September 20, 2024.
- There is no registration fee for this workshop. Breakfast, lunch, and snacks will be generously provided by the Center for Computational Astrophysics (CCA), the Flatiron Institute, and the Simons Foundation, which are hosting the workshop.
- Workshop participants are welcome to stay on Friday for unstructured collaboration time and interaction with CCA members and other workshop participants.
-
Shy Genel, Ph.D.
Flatiron Institute
Francisco Villaescusa-Navarro, Ph.D.
Flatiron Institute
Daniel Anglés-Alcázar, Ph.D.
University of Connecticut
Matt Ho, Ph.D.
IAP / Columbia University
Carolina Cuesta-Lazaro, Ph.D.
CfA / MIT
Boryana Hadzhiyaska, Ph.D.
UC Berkeley
-
Simeon Bird, UC Riverside
Carolina Cuesta-Lazaro, CfA / MIT
Laurence Levasseur, Université de Montréal
Chris Lovell, University of Portsmouth
Daisuke Nagai, Yale University
Yueying Ni, CfA
Paul Torrey, University of Virginia
Benjamin Wandelt, IAP / CCA / JHU -
Events Contact
Cristina Duncan,
[email protected]SOC
Shy Genel, Ph.D.
Francisco Villaescusa-Navarro, Ph.D.
Daniel Anglés-Alcázar, Ph.D.
Matt Ho, Ph.D.
Carolina Cuesta-Lazaro, Ph.D.
Boryana Hadzhiyaska, Ph.D.
-
Speaker Agenda
9:00-9:30 Breakfast
9:30-9:50 Introduction by the SOC
9:50-10:20 Daisuke Nagai (Invited talk): Harnessing CAMELS: A Strategic Approach to Precision Cosmology and Galaxy Astrophysics
10:20-10:40 Erwin Tin-Hay Lau: X-raying CAMELS: Constraining CGM physics with eRASS
10:40-11:00 Bhuvnesh Jain: Higher order statistics for weak lensing
11:00-11:30 Break
11:30-11:50 ChangHoon Hahn: Reconstructing Quasar Spectra and Measuring the Ly$\alpha$ Forest with Spectrum Autoencoders
11:50-12:10 Boryana Hadzhiyska: Detecting large baryonic feedback around DESI photometric galaxies
12:10-12:30 Arne Thomsen: Simulation-based inference of cosmology from multi-probe maps using deep learning
12:30-1:50 Lunch break
1:50-2:20 Simeon Bird (Invited talk): The PRIYA Simulation Suite
2:20-2:40 Alexandra Amon: Connecting astrophysics and cosmology: a weak lensing perspective
2:40-2:50 Biwei Dai: Cosmology from HSC Y3 weak lensing field-level analysis
2:50-3:00 Isa Medlock: Probing Baryonic Feedback and Cosmological Tension with Fast Radio Bursts: Insights from CAMELS simulations
3:00-3:30 Break
3:30-4:10 Lightning poster presentations
4:10-4:20 John Soltis: Direct Estimation of Galaxy Cluster Mass Accretion Rates using Machine Learning
4:20-4:30 Ben Oppenheimer: Novel Multi-Modal Networks to Infer Halo Masses and CGM Gas Fractions
4:30-5:00 Discussion: Nicholas Battaglia and Ana Maria Delgado
5:00-7:00 Welcome reception
-
Speaker Agenda
9:00-9:30 Breakfast
9:30-10:00 Benjamin Wandelt (Invited talk)
10:00-10:20 Lucas Makinen: Interpretable, Scalable Galaxy Formation Parameter Inference at the Population Level
10:20-10:40 Matthew Ho: Practical Cosmological Inference with Learning the Universe
10:40-11:00 Nicolas Cerardi: Lagrangian Deep Learning with CAMELS : towards simulation-based inference for cluster cosmology
11:00-11:30 Break
11:30-11:50 Ben Horowitz: Field Level Inference with Differentiable Hydrodynamical Simulations
11:50-12:10 Aizhan Akhmetzhanova: Detecting Model Misspecification in Cosmology with Multiscale Normalizing Flows
12:10-12:30 Nesar Soorve Ramachandra: Calibrating sub-grid physics in large scale structure simulations
12:30-1:50 Lunch break
1:50-2:20 Carolina Cuesta-Lazaro (Invited talk)
2:20-2:40 Yongseok Jo: Towards Robustness Across Different Cosmological Models
2:40-2:50 Kai Lehman: Learning Optimal and Interpretable Summary Statistics of Cosmological Simulations
2:50-3:00 Jozef Bucko: Leaping into non-Gaussian regime with SBI and Machine Learning
3:00-3:30 Break
3:30-3:40 Serafina Di Gioia: gaBI : a parallel Python package for fast causal discovery from observational data in Astrophysics
3:40-3:50 Eirini Angeloudi: Inferring the 2D ex-situ stellar mass distribution of galaxies with diffusion models
3:50-4:00 Lucia A Perez: Constraints on Cosmology and Astrophysics with Observed Galaxy Clustering: the new photometric expansion to CAMELS-SAM
4:00-4:10 Alice Matthews: Unveiling Galaxy Evolution through UV Luminosity Functions with Simulation-Based Inference: Insights from CAMELS and Illustris-TNG
4:10-4:40 Discussion: Matthew Ho, Carolina Cuesta-Lazaro
4:40-7:00 Poster session and reception
-
Speaker Agenda
9:00-9:30 Breakfast
9:30-9:50 Gregoire Aufort
9:50-10:10 John Wu: Physical insights on galaxy evolution from sparse coding
10:10-10:30 Gabriella Contardo: Cosmology with One Galaxy: No degeneracies is all you need?
10:30-10:40 Julen Expósito: A probabilistic deep learning model to infer DM distribution of dwarf galaxies with HI gas and stellar kinematics
10:40-10:50 Jorge Sarrato Alós: Inferring the dark matter distribution in galaxies with machine learning methods
10:50-11:20 Break
11:20-11:50 Yueying Ni (Invited talk): Harnessing Generative Models in Cosmological Simulations: Applications and Challenges
11:50-12:10 Tjitske Starkenburg: The galaxy physics and dust attenuation in the CAMELS-SAM universe
12:10-12:30 Boon Kiat Oh: CAMELS in details: How cosmological and astrophysical parameters affect galaxy properties
12:30-2:00 Lunch break
2:00-2:30 Paul Torrey (Invited talk): What happens when CAMELS DREAM
2:30-2:50 Jonah Rose: Investigating the Small-Scale Tensions with DREAMS
2:50-3:10 Giulia Despali: The AIDA project: galaxy formation in alternative dark matter models
3:10-3:40 Break
3:40-3:50 Matt Gebhardt: Cosmological backreaction of baryons on dark matter in the CAMELS simulations
3:50-4:00 Weiguang Cui: THE300 – moving towards a better halo mass estimation
4:00-4:10 Xavier Sims:The Impact of Local Environment on Galaxy Evolution in CAMELS
4:10-4:20 Patricia Iglesias-Navarro: Pixel-based fitting of stellar population properties from JADES photometry using Normalizing Flows
4:20-4:30 David Robinson: Connecting galaxy cooling and heating functions to the incident radiation field with machine learning
4:30-5:00 Discussion: Sotiria Fotopoulou and Mikhail Medvedev
-
Speaker Agenda
9:00-9:30 Breakfast
9:30-9:50 Yuri Oku and Kentaro Nagamine: Metal Enrichment in CROCODILE
9:50-10:10 Max E. Lee: Zooming by in the CARPoolGP lane: new CAMELS-TNG simulations of zoomed-in massive halos
10:10-10:30 Sneh Pandya: Learning Galaxy Intrinsic Alignment Correlations
10:30-10:40 Nicholas Frontiere: The Impact of Exascale Simulations
10:40-10:50 Jeger Broxterman: The FLAMINGO hypercube project
10:50-11:00 Megan Tillman: The Effect of Galactic Feedback on the Lyman-alpha Forest Flux Power Spectrum
11:00-11:20 Break
11:20-11:40 Chun-Hao To: Simulations for Current and Future Lensing Surveys: From DES to Roman
11:40-12:10 Chris Lovell (Invited talk): Forward modelling hundreds of millions of galaxies with Synthesizer
12:10-12:40 Laurence Perreault-Levasseur (Invited talk): Data-Driven High-Dimensional Inverse Problems: A Journey Through Strong Lensing Data Analysis
12:40-2:00 Lunch break
2:00-2:20 Shivam Pandey: Accelerated forward models for galaxies and CMB
2:20-2:40 Dhayaa Anbajagane: Map-level baryonification: Efficient modelling of higher-order correlations in the weak lensing and thermal Sunyaev-Zeldovich fields
2:40-3:00 Tri Nguyen: How DREAMS are made: Emulating Satellite Galaxy and Subhalo Populations with Diffusion Models and Point Clouds
3:00-3:30 Break
3:30-3:50 Anshuman Acharya: Machine Learning on simulations and for simulations
3:50-4:00 Florian Kéruzoré: Picasso: a machine learning model to paint intracluster gas on gravity-only simulations
4:00-4:10 Michael Kovac: Simulating the tSZ Effect: Baryonification and Gas Behavior in Galaxy Clusters
4:10-4:20 Supranta Sarma Boruah: Using generative machine learning methods for field-level inference
4:20-4:30 Mauro Rigo: JERALD: high-resolution dark matter and baryonic maps from lower resolution approximate N-body simulations
4:30-5:00 Discussion: Jonathan Blazek and Stephanie Tonnesen
-
Tom Abel
Anshuman Acharya
Prakruth Adari
Aizhan Akhmetzhanova
Juan Pablo Alfonzo
Alexandra Amon
Dhayaa Anbajagane
Eirini Angeloudi
Daniel Anglés-Alcázar
Gregoire Aufort
James Baldwin
Fred Angelo Batan Garcia
Nicholas Battaglia
Simeon Bird
Jonathan Blazek
Supranta S. Boruah
Haley Bowden
Helena Brittain
Jeger Broxterman
Jozef Bucko
Juan Calles
Nicola Cerardi
Alice Chen
Žofia Chrobáková
Gabriella Contardo
Carolina Cuesta-Lazaro
Weiguang Cui
Francis-Yan Cyr-Racine
Biwei Dai
Ana Maria Delgado
Serafina Di Gioia
Julen Expósito
Sotiria Fotopoulou
Nicholas Frontiere
Alex Garcia
Matt Gebhardt
Shy Genel
Boryana Hadzhiyaska
ChangHoon Hahn
Sultan Hassan
Matt Ho
Ben Horowitz
Patricia Iglesias Navarro
Maja Jablonska
Bhuvnesh Jain
Hannah Jhee
Yongseok Jo
Florian Kéruzoré
Kassidy Kollmann
Michael Kovac
Niyantri Krishnan
Erwin Lau
Max E. Lee
Kai Lehman
Shurui Lin
Chris Lovell
Amanda Lue
Lucas Makinen
Alice Matthews
Isa Medlock
Mikhail Medvedev
Jonathan Mercedes Feliz
Jamie Sullivan
Satvik Mishra
Devina Mohan
Daisuke Nagai
Ken Nagamine
Roman Nagy
Aina Nambena
Chloe Neufeld
Tri Nguyen
Yueying Ni
BK Oh
Yuri Oku
Shivam Pandey
Sneh Pandya
Minsu Park
Lucia A Perez
Chad Popik
Alvise Raccanelli
Nesar Ramachandra
Mauro Rigo
Ryan Roberts
David Robinson
Jonah Rose
Christopher J. R. Rowe
Jorge Sarrato Alós
Bryan Scott
Snigdaa Sethuram
Xavier Sims
John Soltis
Tjitske Starkenburg
Richard Stiskalek
Sagan Sutherland
Arne Thomsen
Megan Tillman
Chun-Hao To
Stephanie Tonnesen
Paul Torrey
Sachin Venkatesh
Francisco Villaescusa-Navarro
Ze’ev Vladimir
Benjamin Wandelt
John Wu
Jacky Yip
Sandy Yuan
Kunhao Zhoong
Dhruv Zimmerman
-
In order to foster discussion, the attendees were asked to respond to this poll after a day of talks focusing on large-scale structure cosmology with simulations and machine learning. The results (shown here) were the jumping off points for an open and engaging discussion.