Sam Power
@spmontecarlo.bsky.social
Lecturer in Maths & Stats at Bristol. Interested in probabilistic + numerical computation, statistical modelling + inference. (he / him). Homepage: https://sites.google.com/view/sp-monte-carlo Seminar: https://sites.google.com/view/monte-carlo-semina
Back in my "watching Simons Institute live-streams" era (www.youtube.com/@SimonsInsti...). Very influential on me as a curious PhD student, influential once more as a guy with some free time in the summer!
Simons Institute for the Theory of Computing
The Simons Institute for the Theory of Computing is the world's leading venue for collaborative research in theoretical computer science. Established on July 1, 2012, the Institute is housed in Calvin...
youtube.com
Always interesting to see the different responses to the various items of maths news as they come out. I feel like I find it important to keep reminding myself that even for fairly proximate fields to mine within maths, there are lots of internal cultural aspects on which I'm not well-informed.
Just to give some context: let F be a function that is reasonably continuous but not quite Lipschitz (e.g. Hölder cts). Then, define F_t (x) = sup { F(y) - t || x - y|| }. Then F_t is t-Lipschitz, and pointwise, F(x) < F_t (x) < F(x) + c_t for some constant offset c_t.
Sam Power@spmontecarlo.bsky.social · 7d agoI feel like this phenomenon is somehow quietly omnipresent around convex optimisation; a little bit of differentiability is never too far away, but asking for too much differentiability is a bit dangerous.
Registration is now open for this workshop! website: sites.google.com/view/newcast... registration: webstore.ncl.ac.uk/conferences-...
Non-Equilibrium Sampling: DiffusionsFlowsParticles | Newcastle University WebStore
The Second Sampling Workshop at Newcastle!This three-day workshop will bring together researchers at the forefront of modern sampling techniques and relate
webstore.ncl.ac.uk
Sam Power@spmontecarlo.bsky.social · 2mo agographic design remains my passion sites.google.com/view/newcast...
Lots of interesting bits and pieces cropping up in (what I've encountered of) Tao's ICM talk. I do find that this aspect of 'digestion' of mathematics is generally quite exciting to me, particularly in terms of revisiting established work. (I do understand why it's less lucrative in the short term)
A pair of notes which review and optimise a couple of fun (and by now, reasonably well-established) techniques for showing the 'contraction-on-average' property for stochastic systems, focusing on problems for which the 'obvious' metrics do not contract. (links below)
The semi-answer one can also give here is that many of them were also interested in entropy - though often a different entropy in each case!
Sam Power@spmontecarlo.bsky.social · 2w agoTheir respective areas:
The connection: despite working in dramatically different areas of mathematics, all had sufficient overlap with the work of A.N. Kolmogorov that they contributed to an obituary volume reviewing his contributions to their area.
Sam Power@spmontecarlo.bsky.social · 2w agoa fun trivia: what connects these mathematicians?
(presumably) late to the game, but have been reading a little bit about { Proof Theory / Proof Mining / Skolemization } this afternoon, which has been sort of interesting, especially as someone with a bit of a taste for quantitative results and so on.
Fun to read: djalil.chafai.net/blog/2018/09...
How to give a bad talk - Libres pensées d'un mathématicien ordinaire
Using Beamer. Beamer is a great $\LaTeX$ standard, allowing automation of bad talks. try to maximize the total number of slides, in particular avoid the rule of 2 minutes per slide, and do not hesitat...
djalil.chafai.net
I have just resigned from the board of "Statistics and Computing", along with 17 other Associate Editors. This was a difficult decision: this great journal has published many tremendous articles under the leadership of EiC Ajay Jasra, and before him David Hand, Gilles Celeux, and Mark Girolami.
interesting stuff here: x.com/mehtaab_sawh..., and a bit funny to see the Random Walk Metropolis show up in this context. some inspiration for me to try read it properly and learn something new!
New version available at the same address! We bit the bullet and extended the work to consider arbitrary numbers of blocks, discrete spaces, parallel/systematic/random scan CAVI within a unified framework. Applications now include statistical physics models and others! arxiv.org/abs/2605.30253
Sam Power@spmontecarlo.bsky.social · 2mo agoWith friends at the University of Warwick (in particular, Rocco Caprio and @adriencorenflos.bsky.social), we've recently arXived some work (arxiv.org/abs/2605.30253) on a method for approximate inference known as "Coordinate Ascent Variational Inference", or "CAVI" for short. Let me explain:
Having a nice period at the moment wherein working through the details of basic examples (or examples for basic techniques) points neatly in the direction of new results, either in the sense of suggesting proofs in the general case, or at least in the sense of "it's worth following up on this".
Meaningless grumble: I wish that the "forward process" / "reverse process" terminology for diffusion models had instead been "noising process" and "denoising process". Feels a bit related to how I find ambiguity in the use of "top-down" and "bottom-up" w.r.t. neural networks.
Something which I've realised is very true about my approach to mathematics research is that I'll occasionally see a proof technique ( / assumption / theoretical framework / ... ) so good that I will go out of my way to find a reason to use it for myself. Not sure how prevalent it is (maybe quite?).
The closest I'm getting to home today (stuck at Schiphol on the journey back from Zurich!):
fiddling with the stochastic heat equation (during a talk which is partially about this)
With friends at the University of Warwick (in particular, Rocco Caprio and @adriencorenflos.bsky.social), we've recently arXived some work (arxiv.org/abs/2605.30253) on a method for approximate inference known as "Coordinate Ascent Variational Inference", or "CAVI" for short. Let me explain:
