Periodic Reporting for period 2 - GWmining (Gravitational-wave data mining)
Reporting period: 2023-04-01 to 2024-09-30
Gravitational-wave astronomy is a cutting-edge field within high-energy astrophysics, offering unique insights into extreme objects such as black holes and neutron stars. Unlike traditional telescopes that rely on light, gravitational-wave interferometers observe the direct messenger of gravity. Gravitational-wave detections allow us to infer not only the astrophysical nature of the sources but also the nature of gravity itself. With more and more detections at our disposal, the field of gravitational-wave astronomy is becoming a big-data discipline.
The project "Gravitational-wave data mining" integrates advanced machine learning techniques with the rich data streams from gravitational-wave interferometers LIGO and Virgo. Gravitational-wave inference encompasses two key levels: (i) characterizing individual black-hole mergers and (ii) putting events together to understand the broader population of sources. On both these two levels, analyzing data with powerful predictions and powerful tools is crucial to fully exploit the exciting data at our disposal.
Our project leverages the increasing significance of machine learning and artificial intelligence, which is becoming ubiquitous in our society. We are equipping students and team members with essential skills in this domain, encouraging a critical and conscious use of such powerful tools.
The project "Gravitational-wave data mining" integrates advanced machine learning techniques with the rich data streams from gravitational-wave interferometers LIGO and Virgo. Gravitational-wave inference encompasses two key levels: (i) characterizing individual black-hole mergers and (ii) putting events together to understand the broader population of sources. On both these two levels, analyzing data with powerful predictions and powerful tools is crucial to fully exploit the exciting data at our disposal.
Our project leverages the increasing significance of machine learning and artificial intelligence, which is becoming ubiquitous in our society. We are equipping students and team members with essential skills in this domain, encouraging a critical and conscious use of such powerful tools.
At the single-event level, we concentrated on the impact of eccentricity in binary black-hole systems. This is crucial for discerning their astrophysical formation channels. We have developed a post-Newtonian formalism that incorporates both spin precession and eccentricity with a mixed analytical and numerical approach. This effort culminated in the re-development of the open-source code PRECESSION, which now stands ready for application to gravitational-wave data.
At the population level, we presented foundational work in hierarchical Bayesian analysis which has laid the groundwork for incorporating selection effects, adopting population-informed priors, and stacking information from multiple events in a consistent fashion. In particular, the project has prototyped a machine learning interpolation method, facilitating the analysis of gravitational-wave data by directly leveraging predictions from stellar-physics simulations.
On the astrophysical front, we have explored the implications of repeated black-hole mergers and their potential occurrence among the events observed by LIGO and Virgo. Additionally, we have developed predictive modeling for next-generation ground-based detectors such as the Einstein Telescope and Cosmic Explorer, as well as the LISA space mission.
We hosted a major conference titled "Gravitational-wave populations: what's next?" (July 2023, Milan, Italy) which provided a worldwide hub for researchers to share progress, focus on the current critical questions in the field, and draw a roadmap for future advances.
At the population level, we presented foundational work in hierarchical Bayesian analysis which has laid the groundwork for incorporating selection effects, adopting population-informed priors, and stacking information from multiple events in a consistent fashion. In particular, the project has prototyped a machine learning interpolation method, facilitating the analysis of gravitational-wave data by directly leveraging predictions from stellar-physics simulations.
On the astrophysical front, we have explored the implications of repeated black-hole mergers and their potential occurrence among the events observed by LIGO and Virgo. Additionally, we have developed predictive modeling for next-generation ground-based detectors such as the Einstein Telescope and Cosmic Explorer, as well as the LISA space mission.
We hosted a major conference titled "Gravitational-wave populations: what's next?" (July 2023, Milan, Italy) which provided a worldwide hub for researchers to share progress, focus on the current critical questions in the field, and draw a roadmap for future advances.
Our pioneering work on the interplay between eccentricity and spin precession in black-hole binaries surpasses the current state of the art, offering a new strategy to incorporate both such observables when interpreting gravitational-wave data. Moving forward, we plan to post-process upcoming data from LIGO/Virgo's fourth observing run using our formalism.
The development of hierarchical Bayesian analysis techniques and machine-learning interpolation methods represents a significant leap forward in population studies of black hole mergers. By addressing selection effects and leveraging predictions from stellar-physics simulations, these methodologies are promising advances to extract deeper information from gravitational-wave data.
Our approach needs to be further refined. In particular, we are exploring cutting-edge machine-learning pipelines, particularly focusing on simulation-based inference techniques, which are applicable to various scenarios in gravitational physics. We believe this strategy might solve some of the bottlenecks we encounter in our current implementation which, while exploiting deep learning for the interpolation problem, still relies on traditional sampling techniques.
The development of hierarchical Bayesian analysis techniques and machine-learning interpolation methods represents a significant leap forward in population studies of black hole mergers. By addressing selection effects and leveraging predictions from stellar-physics simulations, these methodologies are promising advances to extract deeper information from gravitational-wave data.
Our approach needs to be further refined. In particular, we are exploring cutting-edge machine-learning pipelines, particularly focusing on simulation-based inference techniques, which are applicable to various scenarios in gravitational physics. We believe this strategy might solve some of the bottlenecks we encounter in our current implementation which, while exploiting deep learning for the interpolation problem, still relies on traditional sampling techniques.