We are hiring!

We are inviting applications for the Chapman Fellowship in Statistics at Imperial College London.

This is a kind of super-postdoc: 3 years contract, excellent working conditions, and candidates are expected to propose an independent research plan.

www.imperial.ac.uk/jobs//search...

congratulations to Paula Cordero Encinar, Francesca Romana Crucinio and O. Deniz Akyildiz (@odakyildiz.bsky.social‬) for winning the 🏆 Best Student Paper Award 🏆 at #UAI2025 with

"Proximal Interacting Particle Langevin Algorithms"

👉 openreview.net/forum?id=rTq...

Here's how the gradient flow for minimizing KL(pi, target) looks under the Fisher-Rao metric. I thought some probability mass would be disappearing on the left and appearing on the right (i.e. teleportation), like a geodesic under the same metric, but I was very wrong... What's the right intuition?

If anyone missed the RSS Gradient Flow workshop back in March, videos are now available...

Late - but better than never.... www.youtube.com/watch?v=QrPy...

Tim Y. J. Wang, Juan Kuntz, O. Deniz Akyildiz
Training Latent Diffusion Models with Interacting Particle Algorithms
https://arxiv.org/abs/2505.12412

Hi, many thanks for interest. It is a bit late to arrange for online registration - but I will investigate if we can record the talks. I'll keep you posted about this if we could do it - so it could be made online later.

This afternoon I was privileged to examine Alex Glyn-Davies, who produced a great thesis on aspects of physics-informed inference for DE models (arxiv.org/abs/2209.15609, arxiv.org/abs/2409.07101 and arxiv.org/abs/2409.06560) and some anomaly detection for good measure doi.org/10.1016/j.pa....

If anyone is thinking of registering for this, I understand that there are still some spaces available, but the room this event is taking place in has a limited capacity which is likely to be reached... so to avoid disappointment it would be safer to register sooner rather than later ;-).

Luca (Martino) once told me (when I said "MCMC does not have weights") that this is incorrect (in his Sicilian style): When you reject in MCMC, you increase the weight of the current sample. Chains do have replicates, can be written like a weighted sample. High rejection rate *is* weight degeneracy.

this warms my heart as a telecom engineer :)

Very happy to announce the 2nd RSS/Turing Workshop on Gradient Flows for Sampling, Inference, and Learning to take place on 24th of March, 2025!

Registration is open now: tinyurl.com/rss-turing-g...

Please see the webpage for topics and speakers. Hope to see you all there!

Correct at the moment we have not set up an online registration option...

Registration is now open for the 2nd RSS/Turing Workshop on Gradient Flows for Sampling, Inference, and Learning at rss.org.uk/training-eve....

Date: Monday 24 March 2025, 10.00AM - 5.00PM
Location: The Alan Turing Institute

This review paper by @guillaume-garrigos.com on SGD-related algorithms is a fantastic resource, offering elegant, self-contained, and concise proofs in a single, accessible reference. arxiv.org/pdf/2301.11235

O. Deniz Akyildiz, Pierre Del Moral, Joaqu\'in Miguez
Gaussian entropic optimal transport: Schr\"odinger bridges and the Sinkhorn algorithm
https://arxiv.org/abs/2412.18432

Indeed, this is a great question. In the averaging setting, this seems to help at various stages of the proof - but I think, in the settings I think of (where slow and fast system has the same drift), one should be able to get away with usual conditions to prove similar stuff...

Splendid stuff; very digestible and usable:

arxiv.org/abs/2412.05239
'Conditions for uniform in time convergence: applications to averaging, numerical discretisations, and mean-field systems'
- Katharina Schuh, Iain Souttar

Great stuff indeed