Matti Vihola
@mattivihola.bsky.social
Professor of Statistics, University of Jyväskylä. Computational statistics, applied probability, Monte Carlo methods, Bayesian inference.
https://iki.fi/mvihola/
https://iki.fi/mvihola/
New material on the O(T log T) coupling algorithms:
• Example where we use unbiased gradients for maximum likelihood estimation
• Generalised coupling algorithm which can handle potentials/weights that depend on current and previous state (coupling of conditional marginal particle filters)
• Example where we use unbiased gradients for maximum likelihood estimation
• Generalised coupling algorithm which can handle potentials/weights that depend on current and previous state (coupling of conditional marginal particle filters)
June 2, 2025 at 6:14 AM
New material on the O(T log T) coupling algorithms:
• Example where we use unbiased gradients for maximum likelihood estimation
• Generalised coupling algorithm which can handle potentials/weights that depend on current and previous state (coupling of conditional marginal particle filters)
• Example where we use unbiased gradients for maximum likelihood estimation
• Generalised coupling algorithm which can handle potentials/weights that depend on current and previous state (coupling of conditional marginal particle filters)
Interesting - we had a similar problem in the diffuse initialisation paper doi.org/10.1007/s112... but you have probably something else in mind.
Conditional particle filters with diffuse initial distributions - Statistics and Computing
Conditional particle filters (CPFs) are powerful smoothing algorithms for general nonlinear/non-Gaussian hidden Markov models. However, CPFs can be inefficient or difficult to apply with diffuse initi...
doi.org
February 12, 2025 at 8:22 PM
Interesting - we had a similar problem in the diffuse initialisation paper doi.org/10.1007/s112... but you have probably something else in mind.
Nice example of a Metropolis algorithm, like the pCN.
February 12, 2025 at 6:51 PM
Nice example of a Metropolis algorithm, like the pCN.
Choosing (X_n,Y_n) such that n^α (Xₙ−x) = Z = n^β (Yₙ−y) for all n suggests that there is little you can get from this assumption?
February 6, 2025 at 5:39 AM
Choosing (X_n,Y_n) such that n^α (Xₙ−x) = Z = n^β (Yₙ−y) for all n suggests that there is little you can get from this assumption?
You can even adapt the update rates: doi.org/10.1214/11-A...
Adaptive Gibbs samplers and related MCMC methods
We consider various versions of adaptive Gibbs and Metropolis-within-Gibbs samplers, which update their selection probabilities (and perhaps also their proposal distributions) on the fly during a run ...
doi.org
February 5, 2025 at 10:59 AM
You can even adapt the update rates: doi.org/10.1214/11-A...