Identifying Differential Equations from Single Observation with Numerical Methods

Join Professor Sung Ha Kang from Georgia Institute of Technology in this DDPS seminar talk where she explores methods for identifying underlying differential equations from a single set of noisy time-dependent data. Learn how the identification process can be formulated as a linear system with feature matrices and coefficient vectors. The talk covers recent research developments including numerical time evolution (IDENT), robust identification methods using successively denoised differentiation, subspace pursuit, and weak form approaches for ODE and PDE recovery. Discover how weak form methods demonstrate robustness against higher levels of noise and higher-order derivatives in underlying equations. The presentation also addresses Group subspace pursuit for varying coefficient cases and Fourier domain applications for identification. Professor Kang, a distinguished Mathematics Professor at Georgia Tech, brings her expertise in mathematical approaches to image processing, numerical methods, and scientific computing to this comprehensive examination of differential equation identification from limited data. This seminar is organized by the libROM team at Lawrence Livermore National Laboratory as part of their Data-Driven Physical Simulation (DDPS) series.