Galaxies SIG Seminar, 8 Sept 2026
AstronomyData Science
THE AI ANGLE
Super-resolving and deblending astronomical survey data using specialized deep learning architecturesAt a NASA Galaxies Science Interest Group seminar, Shoubaneh Hemmati presented deep learning approaches to overcome the trade-offs in coverage, depth, and resolution where classical deconvolution methods plateau. The talk highlights how neural architectures—including generative, diffusion, shape-preserving residual, and physics-informed models—use survey overlaps as empirical priors to super-resolve images and unblend spectral lines. For faculty, this demonstrates how tailoring specialized machine learning architectures to specific physical measurements can scale data enhancement across wide-area astronomical surveys.
THE TEACHING ANGLE
Students can examine why classical deconvolution plateaus on ill-posed inverse problems and analyze how domain-specific requirements dictate whether to deploy generative, residual, or physics-informed neural architectures.Read the original at science.nasa.gov Generate teaching or study materials
More in Astronomy
- AI’s best coding agent fails 60% of the time — and the data backs it upThe New Stack · September 15, 2026
- Observing Earth from orbit is unpredictable and messy: Could 'liquid' AI clear the view?Phys.org — Technology · September 15, 2026