Lightnews — Scholar-powered news

Sam Power @spmontecarlo.bsky.social · 12h

Another one, drafted for the same workshop, but which didn't quite make it - but which I hope will see some use at a future event!

9

Sam Power @spmontecarlo.bsky.social · 2d

It was fun! Mostly tutorials by the physicists for the statisticians; quite a success from that point of view. Recordings here: youtube.com/playlist?lis...

CoSInES-Bayes4Health Masterclass on Computational Physics, April 2024 - YouTube

youtube.com

5

Sam Power @spmontecarlo.bsky.social · 2d

A drafted logo for a past workshop (we ended up going with something else in the end, but I still find this one cute in its own way).

2 21

Reposted by Sam Power

Sam Power @spmontecarlo.bsky.social · 4d

I'll order these so that I can tick them off a list on my computer; the numbers are not any sort of ranking. Summaries will be kept brief and hopefully stoke curiosity, rather than providing answers.

Sam Power @spmontecarlo.bsky.social · 4d

Have been rather quiet on here recently (for uninteresting and unconcerning reasons), but this shouldn't be taken as a lack of enthusiasm for papers which have been coming out recently - loads of really clever and creative works popping up. Will try to post about some of them this afternoon!

2 4 18

Sam Power @spmontecarlo.bsky.social · 3d

This is indeed in the works (after combing through some of my folders of notes), but in the interim, I can share a few things which I've put up directly as .pdf files on my website (sites.google.com/view/sp-mont...), rather than as blog posts per se. Notes 3-5 are 'new'.

8

Sam Power @spmontecarlo.bsky.social · 4d

Glad to hear both!

Sam Power @spmontecarlo.bsky.social · 4d

Regrettable that this reads to me as cognate to "cash my gold".

3

Sam Power @spmontecarlo.bsky.social · 4d

I guess it's really "zero-variance when it should be" (which is already difficult!) rather than "zero-variance in general" (which would be a miracle). A really nice work though; resolves a problem which I'm sure has bothered plenty of people (myself included) for some time.

1 1

Sam Power @spmontecarlo.bsky.social · 4d

Digest to be found here: bsky.app/profile/spmo...

Sam Power @spmontecarlo.bsky.social · 4d

I'll order these so that I can tick them off a list on my computer; the numbers are not any sort of ranking. Summaries will be kept brief and hopefully stoke curiosity, rather than providing answers.

Sam Power @spmontecarlo.bsky.social · 4d

Have been rather quiet on here recently (for uninteresting and unconcerning reasons), but this shouldn't be taken as a lack of enthusiasm for papers which have been coming out recently - loads of really clever and creative works popping up. Will try to post about some of them this afternoon!

Sam Power @spmontecarlo.bsky.social · 4d

(done for now!)

1

Sam Power @spmontecarlo.bsky.social · 4d

27. arxiv.org/abs/2509.20083
'Rethinking player evaluation in sports: Goals above expectation and beyond'
- Robert Bajons, Lucas Kook

The 'expected goals' metric in football omits information about the player taking the shot. How can xG be adapted and extended to evaluate player efficacy?

1 2

Sam Power @spmontecarlo.bsky.social · 4d

26. arxiv.org/abs/2411.07041
'Stochastic parameterisation: the importance of nonlocality and memory'
- Martin T. Brolly

Given a spatially-localised, Markovian-in-time dynamical system, 'optimal' reduced-order modelling can often compromise one or both of these features. (How) shall we recover them?

1 1

Sam Power @spmontecarlo.bsky.social · 4d

25. arxiv.org/abs/2508.10782
'Dimension-Free Bounds for Generalized First-Order Methods via Gaussian Coupling'
- Galen Reeves

Approximate Message-Passing algorithms can be tricky to parse, due to their grounding in large-system asymptotics. A more finitary treatment is both possible and insightful.

Dimension-Free Bounds for Generalized First-Order Methods via Gaussian Coupling

We establish non-asymptotic bounds on the finite-sample behavior of generalized first-order iterative algorithms -- including gradient-based optimization methods and approximate message passing (AMP) ...

arxiv.org

1 1

Sam Power @spmontecarlo.bsky.social · 4d

24. arxiv.org/abs/2510.00389
'Zero variance self-normalized importance sampling via estimating equations'
- Art B. Owen

Even with optimal proposals, achieving zero variance with SNIS-type estimators requires some innovative thinking. This work explains how an optimisation formulation can apply.

2 1

Sam Power @spmontecarlo.bsky.social · 4d

23. arxiv.org/abs/2509.16062
'Transient regime of piecewise deterministic Monte Carlo algorithms'
- Sanket Agrawal, Joris Bierkens, Kengo Kamatani, Gareth O. Roberts

The out-of-equilibrium behaviour of certain MCMC algorithms can exhibit rather different features to the equilibrium behaviour.

1

Sam Power @spmontecarlo.bsky.social · 4d

22. arxiv.org/abs/1910.10667
'Extreme value statistics of correlated random variables: a pedagogical review'
- Satya N. Majumdar, Arnab Pal, Gregory Schehr

Extremes are well-understood in the iid case; weakly-dependent systems are often qualitatively similar. Strong dependence yields new phenomena.

1 1

Sam Power @spmontecarlo.bsky.social · 4d

21. arxiv.org/abs/2402.07048
'Logistic-beta processes for dependent random probabilities with beta marginals'
- Changwoo J. Lee, Alessandro Zito, Huiyan Sang, David B. Dunson

Modelling of spatial dependence becomes challenging outside of the Gaussian setting - could latent Gaussianity suffice?

1 1

Sam Power @spmontecarlo.bsky.social · 4d

20. math.nyu.edu/~kohn/papers/kohn-serfaty-cpam.pdf
'A deterministic-control-based approach to motion by curvature'
- Robert Kohn, Sylvia Serfaty

Can deterministic control problems give rise to HJB PDEs which softly imply the presence of underlying stochasticity?

1 1

Sam Power @spmontecarlo.bsky.social · 4d

19. arxiv.org/abs/2507.14535
'Inference for Diffusion Processes via Controlled Sequential Monte Carlo and Splitting Schemes'
- Shu Huang, Richard G. Everitt, Massimiliano Tamborrino, Adam M. Johansen

How shall we best discretise dynamical systems for the purpose of conducting inference?

1 2

Sam Power @spmontecarlo.bsky.social · 4d

18. arxiv.org/abs/2509.19707
'Diffusion and Flow-based Copulas'
-David Huk, Theodoros Damoulas

If a distribution has standard Gaussian marginals, then one can smoothly interpolate it towards an exact Gaussian along a path which never changes those marginals, only altering the copula.

1 3

Sam Power @spmontecarlo.bsky.social · 4d

17. arxiv.org/abs/1807.00027 + arxiv.org/abs/2508.19648

'Bounds on the Poincaré constant for convolution measures'
- Thomas A. Courtade
'Subadditivity of the log-Sobolev constant on convolutions'
- Thomas A. Courtade, Edric Wang

How does convolution interact with 'strong' concentration of measure?

1 2

Sam Power @spmontecarlo.bsky.social · 4d

16. arxiv.org/abs/2510.01901
'Knots and variance ordering of sequential Monte Carlo algorithms'
- Joshua J Bon, Anthony Lee

Can one ever say that one Feynman-Kac representation of a probability measure is 'strictly better' than another, from an estimation point of view?

1 4

Sam Power @spmontecarlo.bsky.social · 4d

15. arxiv.org/abs/2502.18475
'Least squares variational inference'
- Yvann Le Fay, Nicolas Chopin, Simon Barthelmé

Variational inference is based on a minimisation problem, the stationarity condition for which yields an equation to be solved. Can one attack that stationary condition directly?

1 4

Sam Power @spmontecarlo.bsky.social · 4d

14. arxiv.org/abs/2509.08619
'A hierarchical entropy method for the delocalization of bias in high-dimensional Langevin Monte Carlo'
- Daniel Lacker, Fuzhong Zhou

Bias bounds for unadjusted MCMC methods can often appear pessimistic in practice. Under what circumstances might things be better?

1 3

Sam Power @spmontecarlo.bsky.social · 4d

13. arxiv.org/abs/2510.01933
'Central Path Art'
- Thor Catteau, Benjamin Glancy, Allen Holder, Angela Milkowski, Alexa Renner, Connor Tasik, Rebecca Testa

The interior-point method is effective for solving convex optimisation problems. Can it also be effective for generating art?

1 2