- This event has passed.
Seminar: An ‘Out of the Box’ Approach to Peculiar Velocity Reconstruction using Machine Learning
July 15, 2025 @ 1:00 pm - 2:30 pm
Title:
An ‘Out of the Box’ Approach to Peculiar Velocity Reconstruction using Machine Learning
Speaker:
Rianna Bell, University of Queensland, Australia
Abstract:
Peculiar velocities are some of the most versatile measurements available to cosmologists, having been used over the last 30 years not only to directly constrain key cosmological parameters within the standard ΛCDM model, but also to test the validity of the assumptions and theories that underpin our current understanding of cosmology. One of the most commonly used methods of measuring peculiar velocities is via the reconstruction of the underlying peculiar velocity field using redshift observations. Galaxy redshifts can be used to trace the large-scale structure of the local universe, which in turn can be used to estimate the resulting peculiar velocities of the galaxies. However, it is generally very difficult to obtain an accurate, detailed map of peculiar velocities using this reconstruction process, due to the complex, highly non-linear relation between redshift-space galaxy distribution and the underlying peculiar velocity field.
Given these challenges, machine learning has rapidly become one of the most popular tools for performing these velocity reconstructions. Machine learning models are powerful tools for recognising and extrapolating complex patterns within data. This means that these ML models are able to provide empirical modelling of inverse problems that are otherwise intractable, such as the reconstruction of the peculiar velocity field. However, despite the initial promise shown by convolutional machine learning models in producing more accurate velocity field reconstructions, the detail and accuracy of these reconstructions is inherently limited because these models are not optimised to process the sparse, unstructured redshift observations used for the reconstructions. In this talk, I will present a novel two-step machine learning approach that aims to overcome the limitations of existing machine learning reconstructions in order to provide a more accurate and detailed map of the peculiar velocity field.
In this talk, Rianna will present a novel two-step machine learning approach that overcomes the limitations of traditional and convolutional ML models in handling sparse, unstructured redshift data. This method aims to provide a more accurate and detailed map of peculiar velocities; a key step toward unlocking the full potential of these cosmological measurements.