What does this research mean for the field?

The Frankenfilter, a partially alive particle filter, provides unbiased likelihood estimators for pseudo-marginal Metropolis-Hastings that are more robust to outliers and 2-3 times more efficient than standard particle filters. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

To introduce a partially alive particle filter that prevents infinite simulation loops while generating unbiased likelihood estimators for pseudo-marginal Metropolis-Hastings in hidden Markov models.

September 10, 2026Journal of Computational and Graphical StatisticsOpen Access

Robust, partially alive particle Metropolis-Hastings via The Frankenfilter

View Full Paper

Ask AI

Bookmark

Authors

CSChris SherlockLancaster University AGAndrew GolightlyDurham University ALAnthony LeeUniversity of Bristol

Discussion

Member takes

Implication

Methodological study demonstrates that the Frankenfilter increases robustness to outliers and doubles computational efficiency in hidden Markov models, highlighting its value for Bayesian inference.

Key Points

To introduce a partially alive particle filter that prevents infinite simulation loops while generating unbiased likelihood estimators for pseudo-marginal Metropolis-Hastings in hidden Markov models.
Designed the Frankenfilter algorithm, which fixes explicit lower and upper bounds on simulated particles while pursuing a user-specified target number of non-zero conditional likelihoods.
Embedded the Frankenfilter into a pseudo-marginal Metropolis-Hastings (PMMH) framework to ensure unbiased parameter likelihood estimation.
Evaluated algorithm robustness and runtime performance against standard particle filters in the presence of outlying observations and misspecified starting parameters.
PMMH implemented with the Frankenfilter proved substantially more robust to extreme outliers and poorly chosen initial parameters than standard particle filter baselines.
Computational efficiency improved by a factor of 2 to 3 compared to standard particle filter approaches.
Established an optimal tuning rule for exact observations, proving that targeting n successes is ideal when analyzing n exact observations.

Cite This Study

Sherlock et al. (2026) studied this question.

synapsesocial.com/papers/6aa27a0b58559d80afc72bba https://doi.org/10.1080/10618600.2026.2732158

View Full Paper

Ask AI

Bookmark

Implication

Methodological study demonstrates that the Frankenfilter increases robustness to outliers and doubles computational efficiency in hidden Markov models, highlighting its value for Bayesian inference.

Robust, partially alive particle Metropolis-Hastings via The Frankenfilter

Discussion

Member takes

Implication

Key Points

Cite This Study

Robust, partially alive particle Metropolis-Hastings via The Frankenfilter

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider