This research proposes a machine learning pipeline for real-time detection of photon ring asymmetries in black hole images observed by the Event Horizon Telescope. The pipeline integrates CNNs with physics-informed constraints to identify brightness variations and distortions in black hole shadow images, optimized for real-time inference.
Key findings
Proposes a comprehensive machine learning pipeline for real-time detection of photon ring asymmetries.
Integrates convolutional neural networks with physics-informed constraints to identify azimuthal brightness variations and elliptical distortions.
Employs a multi-scale feature extraction architecture combined with attention mechanisms to capture local and global asymmetry indicators.
Designs a novel asymmetry quantification metric combining morphological features with physical priors from GRMHD simulations.
Optimized for real-time inference with target latency under 100 ms per image to enable responsive analysis during observational campaigns.
Limitations & open questions
The pipeline's performance in real-world scenarios with varying observational conditions is yet to be validated.
The integration of physics-informed constraints may require adjustments as new physical understanding emerges.