Reproduce “VLBInet: Radio Interferometry Data Classification for EHT with Neural Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Joshua Yao-Yu Lin, Dominic W. Pesce, George N. Wong, Ajay Uppili Arasanipalai, Ben S. Prather, Charles F. Gammie
Researchers developed VLBInet, a neural network pipeline that directly classifies black hole accretion states (Magnetically Arrested Disk or Standard and Normal Evolution) from interferometric data products, bypassing image reconstruction. The pipeline achieved over 70% accuracy on synthetic data with realistic noise, but when applied to 2017 Event Horizon Telescope observations of M87*, it did not indicate a statistically strong preference for either state.
Key insights
- Interferometric data products, even when sparsely sampled, contain sufficient information to distinguish between different black hole accretion states.
- Closure quantities demonstrate immunity to station-based complex gain errors, providing a more robust input for classification under realistic noise conditions compared to complex visibilities.
- Neural networks offer a direct, data-driven approach to parameter extraction from VLBI data, complementing traditional image-domain analysis methods by bypassing image reconstruction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.