the trick is to consider spectral densities (densities of states) and use polynomial approximations for them (with a kernel-ignoring twist)

quite happy with this text, give it a read

As a result, you can look how different choices of shift operators and random graph models affect convergence (in terms of the singular value and the actual rate) which is at least exponential

Enjoy!

We show that the convergence rate is governed by the smallest possible singular value of a minor of the composite graph matrix defined by the labeled set. The size of the labeled set then controls the convergence rate (with the dip at the num. of node features)

And you can formally show all this!

I feel like the idiotic dichotomy of the us politics is really rotting people’s brains out. I mean “do different=do like the other guy” is an insane argument to attempt

a wild guess: there will be a legal framework in the us that prevents posting statistics for measles outbreaks

maybe in some form of cdc policy idk

"i'm a fascist and people are mean to me" is certainly a choice for an oped

the in-app barcode scanner when you're in your local bookshop is actually immensely useful, give it a try

Well, they put it back to avoid these comments 🤦

Seems you made them nuke it

🙋

yep
I'm pretty sure it has literally never been reversed for any organisation

You’ve found an old post ;) it was right when CEU was declared undesirable organisation in russia, so any russian citizen getting money from it or doing some work for it would go to prison