Author | Lightnews

Ankit Pensia

@ankitpensia.bsky.social

67 followers 110 following 3 posts

Statistics and ML Theory https://ankitp.net/

ankitp.net

Posts Media Videos Starter Packs

Ankit Pensia @ankitpensia.bsky.social · Feb 5

Why do theory conferences like COLT (& also ICML, JMLR,..) require us to use the 'Lastname et al. (xx)' citation style? Perhaps it's just inertia.

In my opinion, the “alpha” style [L1L2L3xx] is more compact, informative, and fair (giving visibility to all authors, usually ordered alphabetically).

Reposted by Ankit Pensia

The Women In Science Archive @wisarchive.bsky.social · Dec 7

Mathematician Mary Ellen Rudin was born 100 years ago today! One of the great set topologists of her age, her inspired leaps between mathematical realms was the stuff of legend. Here is her story:

tinyurl.com/mrytfn86

#MathSky #WomenInSTEM #HistSci 🧮🧪

Impossible Creatures and How to Make Them: The Topological Legacy of Mathematician Mary Ellen Rudin

There’s a lot to like about plain old, everyday space. No matter where you are, there’s always a way to get to where you need to go, and you can always figure out how long any route from here to there might be. Calculus works in it, and that means we can solve a lot of problems about how stuff moves through it. No matter what scale you exist at, you never lack for neighbours. Yes, space is pretty great, but what, mathematically, is it about normal space that lets it be that way, and what can you

tinyurl.com

2 12 60

Ankit Pensia @ankitpensia.bsky.social · Nov 30

Agreed!

A useful tip: I have recently started using the \citeauthor command for this purpose, and I find it really convenient.

Clément Canonne @ccanonne.github.io · Nov 30

Reminder: if there are 3 authors or fewer, name them all, no "et al."

Actually, try to name all authors if you can, at least the first time you mention the work. And *especially" if the author ordering is alphabetical! "Aaaaa et al." gets old quickly for X, Y, and Z.

1 2

Ankit Pensia @ankitpensia.bsky.social · Nov 26

Not a geometric view but this formula (& its multivariate analog) is useful to show that removing points decreases (appropriately-scaled) variance

We found it useful in the context of designing fast robust algo. that downweigh points iteratively arxiv.org/abs/2204.12399 (here, Bₜ = [E wₜ(X)]² Σₜ)