Sam Duffield

14 Sept

slide prep continues!

29 July

Surely the best lineup in Newcastle since Demba Ba + Papiss Cisse

Sam Power@spmontecarlo.bsky.social

graphic design remains my passion sites.google.com/view/newcast...

13 July

cuthberto's take on the semi-finals please be wrong on Wednesday........ state-space-models.github.io/cuthberto-ca...

github.comGitHub - state-space-models/cuthbert: State-space model inference with JAXState-space model inference with JAX. Contribute to state-space-models/cuthbert development by creating an account on GitHub.

25 June

Wait till he hears about cuthbert github.com/state-space-...

Emily✨@emilyoram.bsky.social

Giroud x Colin 🐛🍫🇫🇷

Sam Duffield reposted

Adrien Corenflos

@adriencorenflos.bsky.social

15 June

I think this didn't get nearly enough traction. Sam did some fantastic work here leveraging the library we've been developing to make predictions for football. This was ported to a UI by Ryan Chan (Warwick MSc student) who's doing his thesis with me (on the methods behind the prediction).

Sam Duffield@samduffield.com

The World Cup is about to start, I wrote a model using cuthbert to predict the games, check out cuthberto-carlos 🐛⚽ ryantjx.github.io/cuthberto-ca...

11 June

The World Cup is about to start, I wrote a model using cuthbert to predict the games, check out cuthberto-carlos 🐛⚽ ryantjx.github.io/cuthberto-ca...

state-space-models.github.ioCuthberto Carlos | World Cup PredictionsInteractive 2026 World Cup match predictions from the Cuthberto Carlos model.

12 June

Now on the state-space-models repo state-space-models.github.io/cuthberto-ca...

sites.google.comNewcastle Non-Equilibrium Sampling WorkshopGroup photo from last year's Newcastle Sampling workshop.

28 May

Co-organising a workshop: Non-Equilibrium Sampling: Diffusions · Flows · Particles September, Newcastle Come join the fun!

21 Apr

I'll be at ICLR this week 🇧🇷 presenting my complete SDE decomposition at the DeLTA workshop on Monday. Reach out if you want to meet!

github.comAdd the Ensemble Kalman Filter by DanWaxman · Pull Request #229 · state-space-models/cuthbertThis PR introduces the ensemble Kalman filter (EnKF). The implementation is based in part on the implementation of CD-Dynamax. That implementation, in turn, is inspired by Algorithm 10.2 in the not...

9 Apr

The ensemble Kalman filter is now in cuthbert 🔥 The EnKF is one of those algorithms that "just works" - oftentimes in settings it has no right to github.com/state-space-...

9 Apr

In this highly nonlinear example it's more accurate and faster than both extended and particle filters. state-space-models.github.io/cuthbert/exa...

state-space-models.github.ioExamples - cuthbertThis section contains examples of how to use cuthbert on some fun problems, highlighting the flexibility of the library and the utility of the underlying methods.

9 Apr

There's a load of fun examples in the cuthbert docs and we're always looking to add more! 🐛 state-space-models.github.io/cuthbert/exa...

18 Feb

It was my birthday this week, it really had to be done

github.com/state-space-...

18 Feb

Sam Duffield reposted

Andrew Gelman et al.

@statmodeling.bsky.social

3 Feb

“Parallelizing MCMC Across the Sequence Length”: This one is really cool. statmodeling.stat.columbia.edu/2026/02/03/p...

statmodeling.stat.columbia.edu “Parallelizing MCMC Across the Sequence Length”: This one is really cool. | Statistical Modeling, Causal Inference, and Social Science

30 Jan

New open source: cuthbert 🐛 State space models with all the hotness: (temporally) parallelisable, JAX, Kalman, SMC

30 Jan

Come check it out if you're interested in time series, Monte Carlo, sequential problems. We've got a suite of fun examples, lots more to add - contributions welcomed! Super fun work with @adriencorenflos.bsky.social and Sahel Iqbal 🙌

30 Jan

Repo: github.com/state-space-...

New preprint! A Complete Decomposition of Stochastic Differential Equations I characterise *all possible SDEs* that satisfy given time-dependent marginals p(x,t)

Here is the decomposition: I show that the scalar ϕ is unique(!) but you can choose and Q or D. In diffusion the ϕ terms represent the "probability flow ODE" but there are actually many ODEs which satisfy p(x,t) depending on your choice of Q

arxiv.orgA Complete Decomposition of Stochastic Differential EquationsWe show that any stochastic differential equation with prescribed time-dependent marginal distributions admits a decomposition into three components: a unique scalar field governing marginal evolution...

Paper for full details. The proofs draw on ideas from vector calculus and Fourier analysis which was really fun to work through arxiv.org/abs/2601.07834

New preprint! A Complete Decomposition of Stochastic Differential Equations I characterise *all possible SDEs* that satisfy given time-dependent marginals p(x,t)

This work combines, unifies and generalises two of my favourite papers - Ma et al - Complete recipe for autonomous SDEs arxiv.org/abs/1506.04696 - Karras et al - Elucidating the Design Space of Diffusion arxiv.org/abs/2206.00364

Sam Duffield reposted

27 Nov 2025

Usual MCMC algorithms are typically guaranteed to work well when used to sample from target distributions for which i) mass is reasonably well-concentrated in the centre of the state space, and ii) the log-density is smooth and of moderate growth. Outside of this setting, things can go poorly.

29 Aug 2025

New paper on arXiv! And I think it's a good'un 😄 Meet the new Lattice Random Walk (LRW) discretisation for SDEs. It’s radically different from traditional methods like Euler-Maruyama (EM) in that each iteration can only move in discrete steps {-δₓ, 0, δₓ}.

29 Aug 2025

As described in the paper, LRW provides multiple benefits but the key motivation for us @normalcomputing.com was the co-design with novel stochastic computing hardware which we believe can drastically accelerate general-purpose SDE sampling.

arxiv.orgLattice Random Walk Discretisations of Stochastic Differential EquationsWe introduce a lattice random walk discretisation scheme for stochastic differential equations (SDEs) that samples binary or ternary increments at each step, suppressing complex drift and diffusion co...

29 Aug 2025

Read more at arxiv.org/abs/2508.20883 Including scaling LRW up to image generation with Stable Diffusion 3.5 🐱

Sam Duffield reposted

13 May 2025

In slides from a recent talk - the { virtuous / vicious } cycle of filtering, smoothing, and parameter estimation in state space models.

27 Apr 2025

Me: Hey so where’s good to eat round here? Singapore taxi driver: Malaysia

27 Apr 2025

Didn’t listen, good decision

18 Apr 2025

posteriors 𝞡 published at ICLR! I’ll be in Singapore next week, let’s chat all things scalable Bayesian learning! 🇸🇬👋

18 Apr 2025

📃 Poster #419 🗓️ Sat 26th, 10:00–12:30 📍 #ICLR2025, Singapore Swing by if you’re into probml, thermodynamic computing or just wanna say hi

18 Apr 2025

We've also updated the paper and made some cool updates to the library 😎 Paper: arxiv.org/abs/2406.00104 Repo: github.com/normal-compu...

Sam Duffield reposted

17 Apr 2025

A new instalment of office decor:

10 Apr 2025

So simple! Normally we order our minibatches like a, b, c, ...., [shuffle], new_a, new_b, new_c, .... but instead, if we do a, b, c, ...., [reverse], ...., c, b, a, [shuffle], new_a, new_b, .... The RMSE of stochastic gradient descent reduces from O(h) to O(h²) arxiv.org/abs/2504.04274

10 Apr 2025

Should have said, here h is stepsize 😅

Sam Duffield reposted

Alex Thiery

@alexxthiery.bsky.social

29 Mar 2025

Sequential Monte Carlo (aka. Particle Socialism?): "why send one explorer when you can send a whole army of clueless one"

Alex Thiery@alexxthiery.bsky.social

Next week is the MCMC chapter of my simulation course. Asked chatgpt to come up with a funny drawing:

24 Mar 2025

Was revisiting the Neural ODEs paper the other day and greatly enjoying. But I found this super confusing, it’s not an A=B+A statement

arxiv.orgNeural Ordinary Differential EquationsWe introduce a new family of deep neural network models. Instead of specifying a discrete sequence of hidden layers, we parameterize the derivative of the hidden state using a neural network. The outp...

24 Mar 2025

arxiv.org/abs/1806.07366

Sam Duffield reposted

sites.google.comSam Power's siteHello! My name is Sam, and I am a researcher in Statistics. I am currently Lecturer in Statistical Science at the University of Bristol. Prior to this role, I was a Senior Research Associate (also at...

20 Mar 2025

Thrillingly (/s), I have today (lightly) updated my website (sites.google.com/view/sp-mont...). I highlight that I've added i) links to several slide decks for talks about my research, and ii) materials related to the (few) short courses which I've given in the past couple of years. Enjoy!

Sam Duffield reposted

AlgoPerf

@algoperf.bsky.social

14 Mar 2025

Hi there! This account will post about the AlgoPerf benchmark and leaderboard updates for faster neural network training via better training algorithms. But let's start with what AlgoPerf is, what we have done so far, and how you can train neural nets ~30% faster.

forum.cursor.comUnexpected indent when using "Run selection in terminal"It seems this issue can be resolved by using the latest version of Python extensions when using Python 3.13. Currently in Cursor, we can only use outdated versions (at least 6 months old) of Python, P...

15 Feb 2025

I've been using Cursor and enjoying it but I'm not sure I'm bullish on it for the long run. One of the best parts about VSCode is the ecosystem of extensions (and that it is open source). Cursor is already out of sync and having issues with extension compatibility

journals.aps.orgThermodynamic matrix exponentials and thermodynamic parallelismThermodynamic computing exploits fluctuations and dissipation in physical systems to efficiently solve various mathematical problems. It was recently shown that certain linear algebra problems can be ...

12 Feb 2025

Thermo Matrix Exponentials has been published in Physical Review Research 🔥 On a thermodynamic computer, the matrix exponential occurs very naturally through the temporal covariance driven by the noise - a polynomial speedup over digital computers!

2 Jan 2025

Discovered today that LinkedIn has daily “Puzzle games” and they’re quite fun!

nature.comThermodynamic linear algebra - npj Unconventional Computingnpj Unconventional Computing - Thermodynamic linear algebra

10 Dec 2024

Thermodynamic Linear Algebra is published! And I, for one, am delighted to see work from Normal Computing published in Unconventional Computing 😝

3 Dec 2024

This is a lovely read

ruiqigao.bsky.social@ruiqigao.bsky.social

A common question nowadays: Which is better, diffusion or flow matching? 🤔 Our answer: They’re two sides of the same coin. We wrote a blog post to show how diffusion models and Gaussian flow matching are equivalent. That’s great: It means you can use them interchangeably.

@adriencorenflos.bsky.social

3 Dec 2024

Although I’d still like to reconcile if FM and diffusion can be seen as optimising the same KL objective with different reparameterisations

Sam Duffield reposted

Adrien Corenflos

24 Nov 2024

Parallel scans accumulate sequences on GPUs (or other parallel hardware) at logarithmic cost in the size of the input. The canonical example is cumulative sums (a, a+b, a+b+c, ...) from an input (a, b, c, ...), but this is hardly the only use, and, e.g., Kalman filtering can be handled in parallel.