InsideOutSMC.jl

This is the companion code for the paper Nesting Particle Filters for Experimental Design in Dynamical Systems, written by Sahel Iqbal and Hany Abdulsamad. It implements $\textrm{Inside-Out SMC}^2$, a nested sequential Monte Carlo algorithm inspired by $\textrm{SMC}^2$ of Chopin et. al. 2013 [1].

Installation

Julia needs to be installed. This has been tested with Julia v1.10. Clone the repo and move to this directory. From the Julia REPL, enter the following commands:

julia> ]
(@v...) pkg> activate .
(@v...) pkg> instantiate

Experiments

The code to reproduce all of the experiments in the paper is given in the experiments directory. All experiments were run on mid-tier laptop CPUs. As an example, to train a policy for the conditionally-linear stochastic pendulum, run experiments/pendulum/linear/trajectory_plot/io_csmc_sysid.jl. If you do not wish to train, a trained policy is already available, and to plot a sample trajectory generated by the policy, run experiments/pendulum/linear/trajectory_plot/make_plot.jl.

Comparison to previous work

In our paper, we compare our method to implicit Deep Adaptive Design (iDAD) from Ivanova et. al. 2021 [2]. The code for that is available here.

Cite

@inproceedings{iqbal2024nesting,
  title = {Nesting Particle Filters for Experimental Design in Dynamical Systems},
  author = {Sahel Iqbal and Adrien Corenflos and Simo S{\"a}rkk{\"a} and Hany Abdulsamad},
  booktitle = {International Conference on Machine Learning},
  year = {2024}
}

References

1. Chopin, N., Jacob, P. E., and Papaspiliopoulos, O. SMC^2: An efficient algorithm for sequential analysis of state space models. Journal of the Royal Statistical Society Series B: Statistical Methodology, 75(3):397–426, 2013.
2. Ivanova, D. R., Foster, A., Kleinegesse, S., Gutmann, M., and Rainforth, T. Implicit deep adaptive design: Policy–based experimental design without likelihoods. In Advances in Neural Information Processing Systems. 2021.