Adaptive Quasi-Monte Carlo Methods for Bayesian Inference and Uncertainty Quantification

-
IIT Tower, 7th Floor conference room
  1. Home
  2. Events
  3. Adaptive Quasi-Monte Carlo Methods for Bayesian Inference and Uncertainty Quantification

Speaker: 

Kan Zhang, Illinois Tech Ph.D. candidate (AMAT)

Description: 

Computing the expected value of a parameter via Bayesian inference involves the numerical approximation of the quotient of two intractable integrals. Traditional Markov Chain Monte Carlo methods suffer from slow convergence. An adaptive quasi-Monte Carlo(QMC) method is proposed to evaluate this quotient to a user-specified error tolerance. The method is illustrated by a logistic regression model and a real-world model. The efficiency of the computation using different methods is studied. When the underlying differential equation contains random coefficients, the quantity of interest involving the solution is random. Compared to standard Monte Carlo, meltilevel methods and quasi-Monte Carlo method may help to reduce the computational cost. The efficiency using different methods is studied.

Event Topic:

Computational Mathematics & Statistics

Getting to Campus

Campus Map & Transportation