New preprint 📑 Fast inference in HMMs with latent Gaussian fields (via SPDE approach + RTMB) ⚡️

🔗 arxiv.org/abs/2603.17469

We modify the forward algorithm to recover a sparse Hessian ➡️ Fast automatic Laplace approximation

Case studies: 1) Detecting stellar flares 2) Lion movement w spatial field

We have a new preprint on covariate-driven #HMMs!
doi.org/10.48550/arX...
@olemole.bsky.social, @rolandlangrock.bsky.social
• commonly used hypothetical stationary distribution can be biased⚠️
• we propose 2 approaches allowing unbiased inference
• simulations and case study on Galápagos tortoises🐢🗺️

Almost! 🤣

Very proud of this paper, where we show that what I've been teaching folks for years is actually really not such a clever thing to do 🙈 But we also provide solutions 💪 Also what a way to kick-start your PhD, @mayavienken.bsky.social 👑

💯

Internally a.k.a. "the JASA paper" since we always felt it should be published in JASA — it really is that good! (JASA folks did not agree 🙈)

Periodically stationary distribution (probability that the fly is active) as a function of the time of day. True stationary distribution is compared to biased approximation, and we see a substantial difference.

Our paper on #HMMs with periodically ⏰ varying transition probabilities is published! 🎉 @carlinafeldmann.bsky.social, Sina Mews, @rmichels.bsky.social @rolandlangrock.bsky.social

doi.org/10.1214/25-AOAS2107

We derive the periodically #stationary distribution and the implied dwell-time distribution

🤣

Priorities! 🥏🥏🥏

And planning this package while playing 🥏 was also great fun 😅

Our review paper on latent Markov models is now published in Statistical Modelling! 🎉 @rolandlangrock.bsky.social @SinaMews.

We discuss choosing the right time and space formulation and provide the R package 📦 LaMa for fast ⚡and flexible estimation.

📄 Paper: journals.sagepub.com/eprint/UETXX...

Fake science!1!! 🤬

Bob, you just need to accept it — everything *is* an HMM 🤷‍♂️

The world is on 🔥 -- and here's my first publication in an astronomy journal: iopscience.iop.org/article/10.3...

We combine Gaussian processes + hidden Markov models to efficiently detect stellar flares in one modelling step. 🧪

How to build your latent Markov model -- the role of time and space Statistical models that involve latent Markovian state processes have become immensely popular tools for analysing time series and other sequential data. However, the plethora of model formulations, t...

Sina Mews, Roland Langrock, and I have updated 🆕 our review paper!
It offers a comprehensive overview on choosing the right time ⏰ and space 📏 formulation for latent Markov models, providing a unifying perspective on discrete- and continuous-time HMMs, SSMs and MMPPs.

👉 arxiv.org/abs/2406.19157