Deep learning-driven likelihood-free parameter inference for 21-cm forest observations
JournalCommunications Physics 8(1), 220 (2025)
First public: Published online:
- Cosmology & gravitation
- Parameter inference
Research summary
The 21-cm forest can probe dark-matter properties and the heating history of the early intergalactic medium. This study uses normalizing flows for simulated-data augmentation and likelihood-free parameter inference. Tests with simulated radio observations examine posterior recovery when the signal distribution is non-Gaussian and simulations are computationally expensive.
Extended notes
Generative flows expand a limited simulation set, while inference flows estimate parameter posteriors from the forest power spectrum. The analysis uses simulated observations relevant to the Square Kilometre Array.