Sequential Monte Carlo style inference support in Stan

Does Stan currently support SMC inference? Especially SMC with different rejuvenation strategies like using HMC?

I couldn’t find anything in the official reference manual. Not sure if this is the right place to ask.


SMC is not currently supported, but I’m tagging Simon (@s.maskell) because I think he and his colleagues are investigating integrating some SMC ideas with Stan. I don’t know enough to give you an update on that work but maybe Simon or one of his collaborators can.

This is definitely the right place to ask!


Yes. We are working to add SMC samplers (with NUTS as the proposal) as a drop-in alternative to MCMC in Stan and with the aim of getting things running on many cored processing environments.

We are also looking at other SMC methoda including particle filters (again with NUTS as a proposal) with the aim of using Stan’s syntax to define dynamic model, using SMC for online learning and at PMCMC (and SMC^2) to learn the parameters of dynamic models.

I’d be fascinated to know if that’s what you meant (!), what applications you had in mind and what motivates the interest.


PS @jonah: thanks.


Congratulations that Liverpool wins the league again after 30 years! SMC is one of my research interests and I am also a crazy Liverpool fan! We all look forward to your SMC samplers in Stan!


Hi, I am also interested in implementing SMC samplers in Stan for static Bayesian models.
I have some ideas in the interest of parallel computation.
May I ask if you are open to cooperation?

Best regards,
Ray Kim

@Red-Portal: yes, we are open to cooperation. We have made quite significant headway already on parallel SMC samplers, but there’s a tonne of work to do so collaborating would be great. Feel free to email me on and we can discuss things in more detail.