Incorporating the Priori into Data-Driven Dynamical Modeling with Deep Learning

Senwei Liang, Purdue University
Supercomputer showdown

Description: With the development of technologies, we have the explosion of time series data from simulation and observation across different fields. Data-driven modeling of dynamical systems provides an important approach for discovering the scientific law behind the data and predicting future phenomena. With superior approximation and generalization capacity, deep neural networks are a prospect for learning the dynamics. Due to the elusive characteristics of the systems, the deep learning approach may have difficulty characterizing the underlying system, and hence integrating the prior knowledge with deep learning is a crucial way to improve the accuracy of the learned dynamics. In this talk, I will introduce my works on dynamics modeling using deep learning with prior knowledge under different scenarios, including prediction with missing dynamics, learning the energy-conserved stiff systems, and estimating the density of time series data.

Join on your computer or mobile app

Click here to join the meeting

Or call in (audio only)

+1 630-556-7958,,419259638#   United States, Big Rock

Phone Conference ID: 419 259 638#