Sparse Orthogonal Regression Technique: A Spectral Framework for Equation Discovery, Approximation, and Integration
SORT learns sparse spectral expansions directly from noisy, irregular data for equation discovery and related estimation tasks.
The paper presents Sparse Orthogonal Regression Technique as an L1-regularized way to estimate orthonormal-basis coefficients without quadrature or analytic inner products. Its main use case is recovering ordinary differential equation vector fields as sparse coefficient expansions. In experiments, SORT matches or improves library-based sparse-regression baselines when the chosen basis fits the problem, and degrades more steadily under sparse sampling, noisy derivatives, or mismatch. The authors frame it as an intermediate representation that can later guide searches for simpler analytic forms. ArXiv · AI/CL/LG's note
The paper presents Sparse Orthogonal Regression Technique as an L1-regularized way to estimate orthonormal-basis coefficients without quadrature or analytic inner products. Its main use case is recovering ordinary differential equation vector fields as sparse coefficient expansions. In experiments, SORT matches or improves library-based sparse-regression baselines when the chosen basis fits the problem, and degrades more steadily under sparse sampling, noisy derivatives, or mismatch. The authors frame it as an intermediate representation that can later guide searches for simpler analytic forms. ArXiv · AI/CL/LG's note
score 5