Linkstream – Telegram
Linkstream
173 subscribers
32 photos
3 videos
2 files
898 links
Various links I find interesting. Mostly hardcore tech :) // by @oleksandr_now. See @notatky for the personal stuff
Download Telegram
ASPLOS'19 // Boosted race trees for low energy classification
Innovation requires constraints, as they say. Here we go, a different machine learning hardware architecture, as performance of von Neumann CPUs isn't improving for years already. Things are getting interesting.
https://sites.cs.ucsb.edu/~sherwood/pubs/ASPLOS-19-racetree.pdf
Genie: A new, fast, and outlier-resistant hierarchical clustering algorithm ('16)
I certainly have a special feeling for non-parametric algorithms. They're kinda fire-and-forget, you plug them in and they work, maybe not as good as with carefully tuned parameters, but you don't need to tune them again and again, which is a big deal.
https://www.gagolewski.com/publications/2016genie.pdf
what phenomena neural net optimizers can and can't generalize?
> Understanding Generalization through Visualizations
https://arxiv.org/pdf/1906.03291v2.pdf
TIL there are mining operations as deep as 3.9 km underground. Also, elevators going 58 kmh!
https://en.wikipedia.org/wiki/TauTona_Mine
OS X <=10.14.5 0day Gatekeeper bypass
TLDR: you can create a ZIP archive with "CoolPicture.JPG.аpp" inside, it will look like JPG and launch without warnings; PoC available too (90day disclosure timeline expired)
https://www.fcvl.net/vulnerabilities/macosx-gatekeeper-bypass
https://tabnine.com/blog/deep -> junior/middle level programmers' salaries down in 3..2..1..
We introduce the heat method for solving the single- or multiple-source shortest path problem on both flat and curved domains. (...) in near-linear time (...) in any dimension, and on any domain that admits a gradient and inner product—including regular grids, triangle meshes, and point clouds. (ACM'17)
https://www.cs.cmu.edu/~kmcrane/Projects/HeatMethod/