François-Xavier Briol

The UCL IMSS Annual Lecture will take place on the 27th April with a keynote from @lestermackey.bsky.social.

The theme is 'Computational Statistics and Machine Learning' and we'll have talks from Alessandro Barp, Paula Cordero Encinar & Po-Ling Loh.

imss2026.github.io

@statisticsucl.bsky.social

Robust goodness-of-fit test with kernels -- now out in JMLR!

jmlr.org/papers/v26/2...

- Existing kernel-based goodness-of-fit tests are not necessarily robust. We explain why in the paper.
- We propose a simple yet provably robust extension.

Joint work with @fxbriol.bsky.social

The recording of my talk on 'Multilevel neural simulation-based inference' at the 'One World Approximate Bayesian Inference' seminar series is now available on YouTube.

Link: www.youtube.com/watch?v=hBWd...

I’ll be giving a talk on a recently accepted NeurIPS paper at the next OWABI seminar on Thursday. The talk will cover simulation-based inference and how you can enhance accuracy when you have cheap approximate simulators at hand.

Thanks Robin! And of course your very nice review of ABC was very helpful (and can be found in the references for those interested!).

Thanks for the interest! I'll send you a separate copy without the breaks via email, but I don't think it'll be better because I tend to make various parts of the slide appear/disappear, so removing breaks will make things look very messy.

Just finished delivering a course on 'Robust and scalable simulation-based inference (SBI)' at Greek Stochastics. This covered an introduction to SBI, open challenges, and some recent contributions from my own group.

The slides are now available here: fxbriol.github.io/pdfs/slides-....

Pre ICML @ London 2025 The goal of this meetup is to bring together students, researchers, and engineers from the greater London area to discuss machine learning research presented at International Conference on Machine Lea...

Today is the day of the pre-ICML event at UCL! Come check out the exciting work from academics, industry researchers, postdocs and PhD students from around London: sites.google.com/view/pre-icm... @statisticsucl.bsky.social @ucl-csml-ellis.bsky.social

I'll give a talk at the BayesComp workshop on 'Bayesian Computation and Inference with Misspecified Models' Tuesday. Come and say "hi!" if you are attending BayesComp!

Workshop website and schedule:
postbayes.github.io/BayesMisspec...

Composite Goodness-of-fit Tests with Kernels, now out in JMLR!

www.jmlr.org/papers/v26/2...

Test if your distribution comes from ✨any✨ member of a parametric family. Comes in MMD and KSD flavours, and with code.

@oscarkey.bsky.social @fxbriol.bsky.social Tamara Fernandez

Pre ICML @ London 2025 The goal of this meetup is to bring together students, researchers, and engineers from the greater London area to discuss machine learning research presented at International Conference on Machine Lea...

Looking forward to the pre-ICML event at UCL on the 3rd July: sites.google.com/view/pre-icm....

Registration and the call for talks/posters are now open!

New review on statistical and machine learning tools to advance equity in genomic research Review led by Dr Brieuc Lehmann and PhD student Leandra Brauninger (both UCL Statistical Science) explores how the analytical methods used to process and interpret genomic data play a critical role in...

New review funded by @genomicsengland.bsky.social & led by Dr Brieuc Lehmann & PhD student Leandra Brauninger (@statisticsucl.bsky.social) explores how the analytical methods used to process & interpret #genomic #data play a critical role in promoting #health #equity: www.ucl.ac.uk/mathematical...

A graphic for the Marie Skłodowska-Curie Actions (MSCA), showing a historical portrait of Marie Skłodowska-Curie overlaid with an image of four young researchers walking down a hallway. The European Commission logo is in the top left. Text reads: "Marie Skłodowska-Curie Actions – €404.3 million to support postdoctoral researchers”

Choose Science. Choose Europe.

A new Marie Skłodowska-Curie Actions Postdoctoral Fellowships 2025 call is now open.

With a budget of €404.3 million, it will support around 1,650 researchers from Europe and beyond.

Apply by 10 September → europa.eu/!fBTMgF

Well done again to @hudsonchen.bsky.social on his very first ICML paper! 😎

Conditional Bayesian Quadrature We propose a novel approach for estimating conditional or parametric expectations in the setting where obtaining samples or evaluating integrands is costly. Through the framework of probabilistic nume...

Surprisingly, @hudsonchen.bsky.social was able to prove a very fast convergence rate! He showed an interpolation rate whereas it was previously believed that only a much slower noisy regression rate was feasible! 🤯

This improves on our prior work: arxiv.org/abs/2406.16530 and all competing methods.

Our paper remedies the problem with a very simple algorithm which is a nesting of two kernel quadrature algorithms. This provably reduces the number of samples needed to obtain a given accuracy when the problem isn't too high dimensional and smooth.

Unfortunately, existing methods such as nested Monte Carlo or multilevel Monte Carlo require a huge number of samples at each level of nesting to estimate these accurately! ☹️

Why should you care? Nested expectations are a significant computational challenge in stats/ML: they arise in active learning, Bayesian optimisation, experimental design, but also other fields such as option pricing and health economics.

New ICML 2025 paper: Nested expectations with kernel quadrature.

We propose an algorithm to estimate nested expectations which provides orders of magnitude improvements in low-to-mid dimensional smooth nested expectations using kernel ridge regression/kernel quadrature.

arxiv.org/abs/2502.18284

Huge congrats to William Laplante (williamlaplante.github.io) on his first paper, completed in the first few months of his PhD!

And thanks to all collaborators including
@maltamiranomontero.bsky.social, Andrew Duncan and Jeremias Knoblauch.

Robust and Conjugate Gaussian Process Regression To enable closed form conditioning, a common assumption in Gaussian process (GP) regression is independent and identically distributed Gaussian observation noise. This strong and simplistic assumption...

The paper builds on the robust and scalable Gaussian process (RCGP) algorithm from our ICML 2024 paper (arxiv.org/abs/2311.00463). It shows that it can make use of common computational tricks in spatio-temporal settings, and also uses adaptive hyper parameter optimisation to improve calibration.

New ICML 2025 paper: Robust Spatio-Temporal GP Regression!

We propose a new GP method with linear-in-time cost that is provably robust to outliers. Unlike competitors, our method is fully conjugate and requires no expensive variational inference!

📄 arxiv.org/abs/2502.02450

If you are at AISTATS this week, check out our paper on cost-aware simulation-based inference!

Cost-aware simulation-based inference Simulation-based inference (SBI) is the preferred framework for estimating parameters of intractable models in science and engineering. A significant challenge in this context is the large computation...

"Cost-aware simulation-based inference" is accepted at AISTATS 2025.

Check out our poster #205 on Sunday May 4th in Hall A-E if you are in Phuket. Finland's rising star @huangdaolang.bsky.social will be there to assist you :D

arxiv.org/abs/2410.07930

@fxbriol.bsky.social @samikaski.bsky.social

Thanks Pierre! I think it could be quite handy for MMD-Bayes and MMD estimators :).

A Dictionary of Closed-Form Kernel Mean Embeddings Kernel mean embeddings -- integrals of a kernel with respect to a probability distribution -- are essential in Bayesian quadrature, but also widely used in other computational tools for numerical inte...

Ever used kernel mean embeddings or MMD? Wished you had a closed-form expression so you wouldn't need to estimate these? See our new paper with Alex Gessner, Toni Karvonen and Maren Mahsereci, which comes with an associated Python package. arxiv.org/abs/2504.18830

So far, 2,135 people have responded to the poll Søren and I posted a few days ago. Of those, 94.4% replied “Yes” to being interested in officially presenting accepted @neuripsconf.bsky.social papers in Europe. (1/7)

We also have excellent line-up of invited speakers which includes @pierrealquier.bsky.social, @eweinstein.bsky.social, Jeremias Knoblauch, David Frazier, Harita Dellaporta, Antonietta Mira, Jeremie Houssineau, Sonia Petrone, Edwin Fong and Aretha Teckentrup.

The BayesComp workshop on 'Bayesian Computation and Inference with Misspecified Models' will take place in Singapore on the 16-17th June.

We have an open call for posters/contributed calls, with a deadline on the 1st May. More details on the website:
postbayes.github.io/BayesMisspec...

JOB OPENING at University of Bristol:
-- Lecturer in Statistics or Machine Learning
-- (x 2 positions available)
-- application deadline: 31 March, 2025

Details at www.bristol.ac.uk/jobs/find/de...

Come join us!