*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Sequential Monte Carlo methods, especially the particle filter (PF) and its various modifications, have been used effectively in dealing with stochastic dynamic systems. The standard PF samples the current state through the underlying state dynamics, then uses current observation to evaluate the samples' importance weight. However, in many applications the current observation provides significant information about the current state while the state dynamics is weak. Sampling using the current observation in this case often produces efficient samples. In this paper, we formulate the framework for a new variant of the particle filter, the independent particle filter (IPF). It generates exchangeable samples of the current state from a sampling distribution that is conditionally independent of the previous states, a special case of which uses only the current observation. Each sample can then be matched with multiple samples of the previous states for evaluating the importance weight. We present some theoretical results showing that this strategy improves efficiency of estimation as well as reduces resampling frequency in many situations. We also discuss some extensions of the IPF. Several synthetic examples and one real example are used to demonstrate the effectiveness of the method.
