TY - GEN
T1 - A Comparison of Clipping Strategies for Importance Sampling
AU - Martino, L.
AU - Elvira, V.
AU - Miguez, J.
AU - Artes-Rodriguez, A.
AU - Djuric, P. M.
N1 - Publisher Copyright: © 2018 IEEE.
PY - 2018/8/29
Y1 - 2018/8/29
N2 - Importance Sampling (IS) methods approximate a targeted distribution with a set of weighted samples, drawn from a proposal distribution. Unfortunately, a mismatch between the proposal and the targeted distribution may endanger the performance of the estimators. In this paper, we focus on the so-called nonlinear IS (NIS) framework, where a nonlinear function is applied to the standard importance weights (IWs). The aim of this transformation is to mitigate the well-known problem of the degeneracy of the IWs by controlling the weight variability. We consider the clipping transformation and test its robustness with respect to the choice of the clipping value. We also propose a novel NIS methodology, where not only a subset of weights is modified a posteriori, but also the corresponding samples are moved. We compare these NIS schemes with standard IS and Monte Carlo methods by means of illustrative numerical examples.
AB - Importance Sampling (IS) methods approximate a targeted distribution with a set of weighted samples, drawn from a proposal distribution. Unfortunately, a mismatch between the proposal and the targeted distribution may endanger the performance of the estimators. In this paper, we focus on the so-called nonlinear IS (NIS) framework, where a nonlinear function is applied to the standard importance weights (IWs). The aim of this transformation is to mitigate the well-known problem of the degeneracy of the IWs by controlling the weight variability. We consider the clipping transformation and test its robustness with respect to the choice of the clipping value. We also propose a novel NIS methodology, where not only a subset of weights is modified a posteriori, but also the corresponding samples are moved. We compare these NIS schemes with standard IS and Monte Carlo methods by means of illustrative numerical examples.
KW - Bayesian Inference
KW - Importance Sampling
KW - Monte Carlo methods
KW - Parameter Estimation
KW - Variance Reduction methods
UR - https://www.scopus.com/pages/publications/85053851334
U2 - 10.1109/SSP.2018.8450722
DO - 10.1109/SSP.2018.8450722
M3 - Conference contribution
SN - 9781538615706
T3 - 2018 IEEE Statistical Signal Processing Workshop, SSP 2018
SP - 125
EP - 129
BT - 2018 IEEE Statistical Signal Processing Workshop, SSP 2018
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 20th IEEE Statistical Signal Processing Workshop, SSP 2018
Y2 - 10 June 2018 through 13 June 2018
ER -