Muldoon, Conor ORCID: https://orcid.org/0000-0003-1381-2561
(2024)
Sensor Placement Algorithms and Learning Methods for Gravitational Wave Interferometers.
In:
Gravitational Wave Science with Machine Learning.
Springer Series in Astrophysics and Cosmology
.
Springer, Singapore.
ISBN 9789819617364 (hardcover); 9789819617371 (ebook)
(In Press)
![]() |
Accepted Version
File not available for download. Available under License In Copyright. Download (629kB) |
Abstract
This chapter explores combinatorial optimization techniques for sensor placementwithinthecontextofapplicationsforgravitationalwavedetectors.Incases where an objective function, being optimised subject to a cardinality constraint, is submodular, or has a diminishing returns property, and is monotone, a simple greedy algorithm achieves near-optimal performance. More broadly, in the context of gravitational wave interferometry, placing sensors requires the optimisation of objective functions that are neither submodular nor supermodular. Such NP-hard optimisation problems can be addressed through the use of a priori hand-crafted heuristics and meta-heuristics inspired by natural, biological, and evolutionary processes. This chapter introduces a novel approach to address the problem by learning latent functions using pointer networks within an actor-critic framework, leveraging attention networks along with deep reinforcement learning. Preliminary implementation challenges and future directions are discussed, highlighting the potential for improved scalability and efficiency for complex optimization tasks.
Impact and Reach
Statistics
Additional statistics for this dataset are available via IRStats2.