All publications

Accelerating Bayesian sampling for massive black hole binaries with prior constraints from conditional variational autoencoder

Hui Sun, He Wang2, Jibo He

  • 2 Corresponding author

JournalPhysical Review D 111(10), 103053 (2025)

First public: Published online:

  • Parameter inference
  • AI for science

Research summary

A conditional variational autoencoder estimates source-parameter distributions for massive-black-hole-binary observations by a space-based detector network. This study then uses the learned distributions to guide the prior range for conventional Bayesian sampling. The comparison examines computational savings alongside differences in posterior shape and the accuracy retained by the combined approach.

Extended notes

The learned model provides a rapid initial estimate, followed by conventional sampling. This combination is evaluated as a way to reduce sampling effort while retaining Bayesian precision.

Sources