Moritz Schauer

Apparently this is my claim to fame

@FreyaHolmer x.com/moritzschauer/…

oh my god I just realized it's called an "equation" because it's got an equals sign that's equating two things

Postdoc in Stochastic modeling of shape ensembles - academicpositions.com/share/twitter/… with Aila Särkkä and me

Applications are open for the CO-OP Master’s 26/27, which is a unique work-integrated programme, combining academic training with industry experience! Hold a Bachelor’s in math, science or engineering & want to help shape a prosperous Africa? Apply via: apply.aims-network.org

Sam Power, Giorgos Vasdekis. [statCO]. Some aspects of robustness in modern Markov Chain Monte Carlo. arxiv.org/abs/2511.21563

A propos…

Neural Guided Diffusion Bridges arxiv.org/abs/2502.11909 w/ @gefanyang @MeulenFrank We introduce a new bridge simulation method that combines the guided proposals of @MeulenFrank and @MoritzSchauer with an additional correction drift term parametrized by a learnable neural network. The family of laws on path space induced by these proposals provides a rich variational family for approximating the law of the diffusion bridge. Once the variational approximation has been learned, independent samples can be generated at a cost similar to that of sampling the unconditioned process. The methods is particularly powerful for conditioning on rare events and for simulating multimodal distributions, which pose challenges for score-learning and MCMC-based approaches.

Sebastiano Grazzi, Giacomo Zanella. [stat.CO]. Parallel computations for Metropolis Markov chains with Picard maps. arxiv.org/abs/2506.09762

Right, you don't need error bars on error bars. Probabilistic uncertainty about uncertainty collapses. This is the “monadic join” in probability. Instead of a coin with random bias p ∼ π, you can flip a coin with the deterministic bias μ. Just take μ = E[p].

We are looking for someone to join the group as a postdoc to help us with scaling implicit transfer operators. If you are interested in this, please reach out to me through email. Include CV, with publications and brief motivational statement. RTs appreciated!

1/4) I am very happy to share our latest work on the information theory of generative diffusion: "Entropic Time Schedulers for Generative Diffusion Models" We find that the conditional entropy offers a natural data-dependent notion of time.

@xuanalogue Same category of mistake as always I want to add

my least favorite thing about RL theory is that it has polluted our understanding of human agency with what is at best a theory of biological agency or brain function

Registration for this years CHAIR Structured Learning Workshop is open. Speakers include: Klaus Robert Müller, Jens Sjölund, @AlexanderTong7, @JanStuehmer @ArnaudDoucet1, Marco Cuturi, @MartaBetcke, Elena Agliari, Beatriz Seoane, Alessandro Ingrosso, ui.ungpd.com/Events/60bfc7b…

@xuanalogue I have to understand it myself first :-) Give me a bit

@MoritzSchauer ooh could you explain?

just Rao-Blackwellized another particle filter and I'm very proud of myself ^_^

@xuanalogue Flow matching, you mean Rao-Blackwellization of the flow??

hire me if you want to Rao-Blackwellize your particle filters, to collapse your Gibbs samplers, to (pseudo-)marginalize out your nuisance variables, etc. etc.

A friend of mine says that Twitter has a dehumanizing effect on a person - it sharpens your wit but hollows out your soul. For some reason, it has the opposite effect on me: my soul feels strangely uplifted, but my sense of self…

Bayesian Inference in the browser? Yup. With new RxInfer TypeScript SDK, enabling real-time, client-side probabilistic reasoning. Think: adaptive UIs, privacy-first personalization and more. Interested in bringing Bayesian intelligence to the frontend? Let's talk. #RxInfer #WebAI

@gill1109 For a spiritual background have a look at the essay ams.org/journals/notic… An optic is very similar to the lenses considered there

@gill1109 Yes, “monoidal functor” tells people from category theory that implementing Bayes rule or BFFG for a hierarchical model is a job for a computer to the extend that if a computer program can sample a generative model, it can also be transformed to perform inference for the model

A paper that started with a tweet, now its submitted: Compositionality in algorithms for smoothing / Moritz Schauer, Frank van der Meulen, Andi Q. Wang arxiv.org/abs/2303.13865

Check out cool new work from our group in collaboration with Pfizer and AstraZeneca, lead by Julian Cremer and Ross Irwin on FLOWR, a flow-based ligand generation approach, and highly sanitized benchmark dataset, SPINDR, for the SBDD community!