I'm trying to understand this paper but I can't figure out what the difference between SIR and SMC is. I thought that SIR is an example of SMC but the authors seem to distinguish between them. They state:

In this section, we show how it is possible to use any local move—including MCMC moves— in the SIS framework while circumventing the calculation of distribution (9),

where (9) corresponds to the importance distribution. However, I don't see why this would be a problem in SIS and how SMC is different. I would be very grateful for help!


A standard reference on this topic is:

A tutorial on particle filtering and smoothing: Fifteen years later. Arnaud Doucet, Adam M Johansen

In particular:

enter image description here

So, as you can see, SMC = SIS + resampling, also known as SIR or SIS/R. This definition is also in the paragraph immediately after equation (1) in the paper your cited. The resampling step, as discussed in section 2.4 of the paper you cited, circumvents the need for calculating the importance distribution in SIS (the equation (9) you refer to), which is impossible to compute is most cases.

| cite | improve this answer | |
  • $\begingroup$ Thank you. To sum up, SMC is the same as SIR. I think this is a reason why I found this article confusing because the authors use both notions. $\endgroup$ – Paula Aug 7 '16 at 11:15
  • $\begingroup$ @Paula Indeed. SMC is a relatively young area, and some concepts and slang have been shaped recently. This is one of the reasons why the tutorial I posted is so popular (~1000 citations) since it puts together several ideas. $\endgroup$ – Goro Aug 7 '16 at 11:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.