Deep learning based reduced order modeling of seismogram-type acceleration time series model: Part - II

Abstract

The Bayesian inference framework with the obvious invocation of Markov Chain Monte Carlo is computationally infeasible if the forward model is heavy. The reason is that the inference framework requires a huge number of forward simulation passes to arrive at robust estimates of the model parameter(s). In lieu of that, a reduced order model becomes critical. The LSTM based encoder decoder framework offers promise in that realm. The encoder effectively compresses all the solution information in dominant eigenmodes, and the decoder reconstructs the solution from those eigenmodes.