Abstract :
Monte Carlo methods are important tools for scientific research, ranging from studying molecular systems in physics and chemistry to learning state space models and randomfields used in speech and language processing.
In this talk, we discuss several interrelated topics:
(i) Monte Carlo algorithms for generating samples from multiple distributions (possibly of different dimensions),
(ii) stochastic approximation for online estimation of normalizing constants during the simulation process,
(iii) stochastic approximation for simultaneous estimation of model parameters and normalizing constants for random fields (possibly of different dimensions),
(iv) global and local weighted histogram analysis methods for offline estimation of normalizing constants and expectations upon completion of the simulation process, and (v) stochastic solution for weighted histogram analysis dealing with a large number (hundreds or more) of distributions.
Biography :
Zhiqiang Tan is a Professor with the Department of Statistics & Biostatistics at Rutgers University. He received the B.S. degree in applied mathematics from Tsinghua University and the Ph.D. degree in statistics from the University of Chicago. He has published extensively on statistical theory, methods, and applications in leading statistical and interdisciplinary journals and conferences. His research interests include Monte Carlo methods, causal inference, and statistical learning.
Supported by the Tsinghua Global Scholars Fellowship Program
