Stefan Güttel

We are playing the awesome @AfterAllFest this weekend!!!! Friends old and new come and join the party @thepeerhat on Sat night (on stage at 9). Buzzin to share a stage with so many awesome acts. Get your tickets here 👇 fatsoma.com/e/848wbm7y/la/… See you down the front 🌠🎶🐝💫

@SIAM_UKIE @LeedsUniComp Congrats Max!

SIAM-UKIE Section warmly welcomes new Secretary & Treasurer, Dr Massimiliano Fasi @LeedsUniComp eps.leeds.ac.uk/computing/staf… Thank you to everyone who voted 🗳️

'Fast and exact fixed-radius nearest neighbor search based on sorting' now available @PeerJCompSci peerj.com/articles/cs-19…

Mathematicians use AI to identify emerging COVID-19 variants manchester.ac.uk/discover/news/…

@reidatcheson Chebyshev interpolation of a step function?

any guess on what I'm about to try and do

@reidatcheson One can sometimes do a little better by selecting the k vectors to orthogonalize against from *all* previous basis vectors, not just the latest ones arxiv.org/abs/2306.03592 But the cond number will grow exponentially eventually for nonsymmetric A (see the thm).

It's a known fact that there aren't stable 3-term recurrences for orthogonalizing Krylov vectors for symmetric matrices, even though in exact precision the Lanczos recurrence does the job. But what if you orthogonalize against a fixed window of past vectors (more than 3?)

I gave an informal talk on the CLASSIX clustering algorithm in our @nla_grp seminar yesterday. It contains some MATLAB and Python live demos. youtube.com/watch?v=Q-CQBE…

Image segmentation in @MATLAB using CLASSIX link.medium.com/CAc0OnLjsFb

Yo 2024 !! We're back Tickets - fatso.ma/Y3J5 #newmusic #inego

Clustering more than 5.7 million data points in less than 8 seconds with CLASSIX (when HDBSCAN would take an hour) link.medium.com/4TewjS4V5Eb

@MarcelSchweitz1 Nice work! We just posted arxiv.org/abs/2311.14206 -- now I wonder how your algorithm would compare to using sketched GMRES for solving the linear system formulation of the matrix equation (with a Kronecker structured sketch operator)

New preprint on randomized Krylov methods for large-scale Sylvester/Lyapunov equations (algorithmic aspects + theoretical analysis). Turns out to be highly competitive with state-of-the-art rational Krylov both regarding run time and memory consumption: arxiv.org/abs/2311.16019

New work on combining sketching and Krylov subspace recycling with Liam Burke from @MathsTCD and Stefan Güttel @guettel. New algorithm GMRES_SDR and theoretical analysis of sketched GMRES, which we build upon to analyse our new method. arxiv.org/abs/2311.14206

CLASSIX a new clustering method that provides textual and visual explanations of its outputs. It is also super fast. It clusters 2 million data points in under a second on a laptop where DBSCAN would require an hour. #scaling-test-classix-vs-dbscan" target="_blank" rel="nofollow noopener">github.com/nla-group/clas…

Happy to announce the release of the CLASSIX #clustering toolbox for @MATLAB github.com/nla-group/clas…

@latzplacian I read "Counterexamples in Analysis" while studying Wegert's 4-semester Analysis course at Freiberg. These counterexamples really annoyed me and I much prefer Complex Analysis now 😅

@OdedRechavi Pilot

Name a job that allows you to travel and see the world more than being in academia

@Math_World_ 101 and many more (1 second)

Which number is missing? In how many seconds did you find?

New work on combining Krylov subspace recycling with randomized sketching for matrix functions and linear systems - arxiv.org/abs/2308.02290 (with Liam Burke @tcddublin)

Our apartment in Varna, Bulgaria, is right above the @twitter headquarters

@CompSysMed @FrancoisFages @_CIRM Any chance that slides and/or video of this talk could be made available please? Thanks