Advancements in Interferometric Imaging Techniques Using Machine Learning
Recent research by Nithyanandan Thyagarajan, Lucas Hoefs, and O. Ivy Wong presents advancements in interferometric imaging techniques through their paper titled "Interferometric Image Reconstruction using Closure Invariants and Machine Learning" (arXiv:2311.06349). The study focuses on the use of interferometric closure invariants, which capture calibration-independent details of an object's morphology, to enhance the classification and parameter estimation of various astronomical shapes.
The authors explored six different morphological models: point-like, uniform circular disc, crescent, dual disc, crescent with elliptical accretion disc, and crescent with double jet lobes. They applied machine learning models, including logistic regression, multi-layer perceptron, and random forest, to analyze closure invariants derived from sparsely covered apertures. The findings indicate that all methods, except logistic regression, achieved classification accuracies exceeding 80%, with improved performance correlating with increased aperture coverage.
In addition to classification, the study also estimated parameters for specific morphologies using multi-layer perceptron models, successfully reconstructing images that correspond well with the input data. However, the accuracy of these reconstructions diminished in cases where parameter degeneracies were present.
This research is particularly relevant for interferometric imaging under challenging observational conditions, such as those encountered by the Event Horizon Telescope and Very Long Baseline Interferometry. The techniques developed in this study could complement existing methods, providing more robust constraints on the morphology of astronomical objects, thereby enhancing our understanding of the universe.
The paper is published in the RAS Techniques and Instruments (RASTI) special edition on "Next-Generation Interferometric Image Reconstruction" and spans 17 pages, including numerous figures illustrating the findings.