Deep Learning Reconstructs Non-Spherical Transit Geometry
A new arXiv study by Ushasi Bhowmick and Shivam Kumaran demonstrates that deep neural networks can extract low-order geometric information about non-spherical objects from stellar transit light curves. The authors generate a large library of two-dimensional random shapes, decompose each into a series of elliptical components using Fourier coefficients, and simulate transits with the Yuti light-curve simulator. Neural networks trained to regress the coefficient set recover the dominant elliptical components reliably, including overall size, orientation, and large-scale perturbations. Higher-order components are only partially recoverable: scale is inferred while eccentricity and orientation become ambiguous. The work quantifies how orientation and non-convex features limit identifiability and frames transit inversion as an intrinsically ill-posed, yet practically informative, inverse problem.
What happened
The arXiv paper 2509.14875, revised on 10 Apr 2026, shows that deep neural networks can learn to map transit light curves to geometric parameters of non-spherical occulting objects. Authors Ushasi Bhowmick and Shivam Kumaran build a synthetic pipeline that generates thousands of two-dimensional random shapes, expresses each shape as a sum of elliptical modes, simulates transits with the Yuti light-curve simulator, and trains networks to predict the modal coefficients.
Scoring Rationale
This is a notable methodological advance for inverse problems in astrophysics and demonstrates practical limits of information extraction from light curves. It is a focused arXiv contribution with clear implications for exoplanet and object-shape inference, so it rates as notable but not industry-shaking.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problemsStep-by-step roadmaps from zero to job-ready — curated courses, salary data, and the exact learning order that gets you hired.
