Interacting Particle Markov Chain Monte Carlo
Authors: Tom Rainforth, Christian Naesseth, Fredrik Lindsten, Brooks Paige, Jan-Willem Vandemeent, Arnaud Doucet, Frank Wood
ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present empirical results that show significant improvements in mixing rates relative to both noninteracting PMCMC samplers and a single PMCMC sampler with an equivalent memory and computational budget. An additional advantage of the i PMCMC method is that it is suitable for distributed and multi-core architectures. |
| Researcher Affiliation | Academia | 1 The University of Oxford, Oxford, United Kingdom 2 Link oping University, Link oping, Sweden 3 Uppsala University, Uppsala, Sweden |
| Pseudocode | Yes | Algorithm 1 Sequential Monte Carlo (all for i = 1, . . . , N) Algorithm 2 Conditional sequential Monte Carlo Algorithm 3 i PMCMC sampler |
| Open Source Code | Yes | An implementation of i PMCMC is provided in the probabilistic programming system Anglican1 (Wood et al., 2014), whilst illustrative MATLAB code, similar to that used for the experiments, is also provided2. 2https://bitbucket.org/twgr/ipmcmc |
| Open Datasets | No | The paper states that '10 different synthetic datasets of length T = 50 were generated by simulating from (12a) (12c)' for the LGSSM and refers to the NLSSM as 'considered by, among others, Gordon et al. (1993); Andrieu et al. (2010)'. While the data generation process for LGSSM is described, there is no concrete access information (link, DOI, repository) provided for the specific synthetic datasets used in the experiments, nor for the NLSSM experimental data. |
| Dataset Splits | No | The paper does not provide explicit training, validation, or test dataset splits (e.g., percentages, sample counts) for its experiments. It focuses on the properties of the MCMC chains and the models rather than data partitioning. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments (e.g., GPU/CPU models, memory specifications). |
| Software Dependencies | No | The paper mentions that an 'implementation of i PMCMC is provided in the probabilistic programming system Anglican' and 'illustrative MATLAB code'. However, it does not specify version numbers for either Anglican or MATLAB, which are necessary for reproducibility. |
| Experiment Setup | Yes | M = 32 nodes and N = 100 particles were used unless otherwise stated. Initialization of the retained particles for i PMCMC and m PG was done by using standard SMC sweeps. We set µ = [0, 1, 1]T , V = 0.1 I, Ω= I and Σ = 0.1 I where I represents the identity matrix. The constant transition matrix, α, corresponds to successively applying rotations of 7π 10 and π 20 about the first, second and third dimensions of xt 1 respectively followed by a scaling of 0.99 to ensure that the dynamics remain stable. |