Bayesian Inference on Structure Identification

EasyChair Preprint 1558

Abstract

We discuss the two specific applications of Bayesian Inference[5] with deep learning[4] specifically in Structure identification problems. As deterministic modeling does not consider uncertainty during computation, it is complementary to combine it with stochastic method, for instance, Bayesian Inference when solving problem requires prior information. In the first problem, MCMC[14] is applied in randomizing the sampling of seismic waves.It is remarkably efficient when the problem transfer from 1D to 2D since the parrellel initialization with different MC Markov Chains do make the search of reconstruction of the seismic faster(getting more chances in validating the prior) and more accurate(less inference bias misleading the inversion and less disturbed by noisy and unrelated information.) In the second problem. t-SNE[17] is conducted with modified Bayesian Information Criteria(BIC) which not only makes the computation easier and faster but also efficiently reduce the high dimensional of the data. The choice of the perplexity is also done automatically which can be updated with every iteration

Keyphrases: Bayesian Variant Inference, Bayesian inference, Interior structure, MCMC(M-H), deep learning, structure identification, t-SNE

@booklet{EasyChair:1558,
  author    = {He Qin},
  title     = {Bayesian Inference on Structure Identification},
  howpublished = {EasyChair Preprint 1558},
  year      = {EasyChair, 2019}}