Shubhendu Trivedi

Wasserstein geometry = quotient geometry of permutation invariance.

In this blog, I explain why this is the natural language for exchangeable particles—and why mean-field neural network training shows up as a W2 gradient flow.

mufan-li.github.io/OT2/

Interestingly never heard of it before today.

Standard sort of pol sci conference. Ivan Krastev is good and very insightful and tends to invite all sorts of people. But most of the people assembled aren't all that good IMO, surprising given he's the curator. If he was looking for a fascist speaker, there could be better options.

I was just pulling your leg!

YouTube video by afterImages 600 Years in Kerala | The Gujarati Community of Calicut | Documentary | afterImages

"Today is ours". Mr. Padamsi on the Gujaratis of Calicut.
www.youtube.com/watch?v=UcFi...

You fought with me 13 years ago for taking a crack at our common crackpot. :P

I asked chatppt a variation and it tries to act too smart

Where is Claude code for packing up a few thousand books when you need it?

I was curious if Obsidian was an actual Armenian surname. Turns out not originally, but after some corruptions and misspellings of some mysterious original name, it exists as an ultra rare surname.

Oh wow! had no idea

I can't even imagine what 61% would have looked like there.

My most successful % wise fluke has a record here. This % gain was on an 8% stock decline. bsky.app/profile/shub...

15-20x is a conservative estimate btw. Since linkedin no longer has a stock I don't know what it's implied volatility numbers were. I also assume it had a thinly traded options chain, so had more slippage. 61% on something as high volume as nvda or tsla would mean 100x even.

Also see how Jim Simons dealt with him. Also Taleb (whatever his flaws, he was skilled at options). Simons differed from these math people because he didn't need to be impressed by money, was a math heavy weight, but also a pretty tough guy (less room for politesse).

There is a record of 800 shares bought for his charity when the linkedin technicals looked awful. In 15d there was news of acquisition, and that closed with a 61% gain in no time. Now imagine if he bought options in personal accounts for 1 MM. Could have easily been 15-20x (based on 61% stock gain).

I am actually not a fan of the Newman stuff. They just kept writing the same thing again and again. But between centrality could be a measure yes. But you could keep it simpler, and use something like Adamic-Adar.

For sparse graphs, we can still do reasonable stuff. We don't want wrong edges to dominate. Could have a first filtering step where you in fact sparsify further based on mutual similarities, local structure (keep locally meaningful edges). Then collapse nodes betwn heavy edges in an iterated manner.

^if I understood you correctly. You don't seem to mean looking at a "reduced graph" where subgraphs are reduced to nodes and the reduced graph is clustered instead and then the clustering is projected up to the full graph.

Yes, it makes sense. Sounds more like personalized page rank and / or seeded clustering. There used to be work using Lancozs / Arnoldi for seeded spectral clustering.

<authors>, <title>, obv

Since bibliographies were/are in the conversation recently, a random tip on bib health: If you cite an arXiv paper with copy paste from some default channels, which renders in references as <title>, <year>, <arxiv url>, it will very rarely get picked up by google scholar.

What do you mean by locally invert?

Not at all. The people who were good at it stayed far away and thought it was mafia money. He was more of a wheeler-dealer PE corporate raider guy. Also, options are easy (w outsized returns due to the leverage) if you have insider information (check his trade in LinkedIn before acquisition by MS).

Prof. P. P. Divakaran (1936-2025) passed away last year. An appreciation by those who knew him.
bhavana.org.in/puthan-puray...

I like that it's something that people enjoy. :)

I will share a list. Vovk doesn't write in an accessible way IMO. It's part of the reason why the overhype/popularity of the framework came due to papers on it by others (Larry Wasserman and Jing Lei at first, then Tibshirani and Barber, and then Candès and Jordan).

Usually learn of super bowl (on some years) because I might go to some of my usual hangs, and they are filled with loud people (a bit out of character for said places). I want to say that I hate this shit, but that's too mean. In some ways, it's "better" than the spectacle around cricket matches.

No shame in trying to copy too e.g. x.com/_onionesque/...

For example, anyone who knows me and talks to me about conformal prediction, always is recommended the same six papers. None of them has Vovk as an author.

Every once in a while I will come across an arxiv submission with a particular author, and I think OK, I will get to learn about this area, even if I don't understand their result/paper. Rina Barber is one such author.