Sequential monte carlo samplers
About
This paper shows how one can use Sequential Monte Carlo methods to perform what is typically done using Markov chain Monte Carlo methods. This leads to a general class of principled integration and genetic type optimization methods based on interacting particle systems.
Pierre Del Moral, Arnaud Doucet• 2002
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Unconditional modeling | Funnel d = 10 | Delta log Z0.561 | 30 | |
| Unconditional modeling | 25GMM d = 2 | Delta Log Z0.569 | 30 | |
| Unconditional modeling | Manywell d = 32 | Δ log Z14.99 | 29 | |
| Target Distribution Sampling | Funnel 10D | Sinkhorn Distance149.3 | 29 | |
| Toy target distribution sampling | GMM40 d = 50 | W2 (Entropy Regulated, eps=0.05)111.8 | 18 | |
| Sampling on discretised synthetic densities | Manywell d = 32 | Sinkhorn Dist.29.14 | 15 | |
| Sampling from synthetic distributions | 25GMM d = 2 | Delta Log Partition Function Error (Zr)0.345 | 13 | |
| Sampling from synthetic distributions | Manywell d = 32 | Partition Function Error (Zr)30.17 | 13 | |
| Learning Continuous Target Distributions | MoS d = 50 | Sinkhorn Cost3.30e+3 | 11 | |
| Target Distribution Sampling | Many-Well 5D | Sinkhorn Distance20.71 | 11 |
Showing 10 of 20 rows