Deep learning for parameter estimation

Speaker:  Caroline Tatsuoka (OSU)

Dates: 2023/10/05

Location: MA105

Abstract: We present methods to obtain estimates to unknown parameters of dynamical systems via data driven methods using deep neural networks (DNNs). An inverse map from the solution space to the parameter space is approximated via DNNs for parameter estimation, free of prior assumptions on the model and using partial data observations of the full system. We present the numerical framework and several examples that demonstrate our ability to recover parameters via noise-free and noisy data observations. Joint work with Zhongshu Xu, Prof. Victor Churchill, and Prof. Dongbin Xiu.