5 ms·

My Favorite Papers of 2017

152 points by snickmy 9y ago

retrievergias 9y ago
TFX is indeed quite a massive one. Well played Google
arvinder 9y ago
Website broken with Firefox on Android.
jacksmith21006 9y ago
My favorite is the Jeff Dean paper from NIPS in using NN for data base indexes.
- snickmy 9y ago
  You are referring to this http://learningsys.org/nips17/assets/slides/dean-nips17.pdf http://learningsys.org/nips17/assets/slides/dean-nips17.pdf, right? It didn't make the cut just because is a keynote and not a paper, but nevertheless is a great engineering fit. Even more if considering that in the meanwhile Jeff is running the entire brain team, and keep pouring contributions on a regular basis: https://research.google.com/pubs/jeff.html https://research.google.com/pubs/jeff.html
  - stablemap 9y ago
    There is a paper for the index structure part; several authors. https://arxiv.org/abs/1712.01208 https://arxiv.org/abs/1712.01208
    - snickmy 9y ago
      Ah, great catch! thanks for sharing
codepie 9y ago
Slightly OT, but I wanted to know if there's any group/forum where we can read and discuss papers regularly? I know the morning paper (https://blog.acolyer.org/ https://blog.acolyer.org/), but it is not so interactive, in the sense that you can't have discussions with peers and it turns into just passive reading.
- snickmy 9y ago
  At Spotify, we have a reading group. I started one at SoundCloud and at Google there were multiple focused on very specific topics. If you can prove to your company the value in investing in a reading group, you can probably start one in your company too. KTH, Stockholm's main technical university, has an open reading group too, open to everyone. Probably if you have a big enough university in your area they have something similar. Alternatively try reddit, ie.: on r/machinelearning. That said, the value of doing this in real life, in front of a whiteboard, is worth going extra mile and find someone to team up with.
- almata 9y ago
  Not exactly what you're looking for, but maybe you or someone else might be interested on this: http://fermatslibrary.com http://fermatslibrary.com
  - shalabhc 9y ago
    Very interesting, thanks! Somewhat related,a general purpose annotation platform is http://hypothes.is http://hypothes.is - a great feature is that PDF annotations are location independent because they use some kind of fingerprint to identify the document.
- dilap 9y ago
  Yeah, I'd be interested in this too. Something with a lot of discussion, smallish group, to create a sense of social obligation... (So easy to be lazy w/ the infinite world of distractions available.)
- AlexCoventry 9y ago
  If you're in Boston, we have a reading group for machine learning papers. We'll start up again on the 9th. List of papers we've discussed: https://github.com/pmiller10/cambridge-ai https://github.com/pmiller10/cambridge-ai Mailing list: http://mailman.mit.edu/mailman/listinfo/sipb-deeplearning http://mailman.mit.edu/mailman/listinfo/sipb-deeplearning
yesenadam 9y ago
Could I slip in a request here for contributions to my Ask of 3 days ago, still on top page, but not many answers yet, What are the best scholarly papers you read in 2017? I'm hoping for some discussion on there. https://news.ycombinator.com/item?id=15991373 https://news.ycombinator.com/item?id=15991373
greenpizza13 9y ago
Huge, massive, credibility-undercutting typo in the slot pulling analogy: "you loose." I'm not sure when people started making this mistake, but it's one of my more cringe-worthy ones to see.
- thomasfoster96 9y ago
  You find a rather common spelling mistake (which has been fixed) “credibility-undercutting” and “cringeworthy”? It seems like you are holding blog posts that have made it to the first page of Hacker News to a ridiculously high standard.
tommyhansen 9y ago
I wish Apple was sharing even more, and not just on Machine learning.

152 points by snickmy 9y ago

retrievergias 9y ago

TFX is indeed quite a massive one. Well played Google

arvinder 9y ago

Website broken with Firefox on Android.

jacksmith21006 9y ago

My favorite is the Jeff Dean paper from NIPS in using NN for data base indexes.

snickmy 9y ago
You are referring to this http://learningsys.org/nips17/assets/slides/dean-nips17.pdf http://learningsys.org/nips17/assets/slides/dean-nips17.pdf, right? It didn't make the cut just because is a keynote and not a paper, but nevertheless is a great engineering fit. Even more if considering that in the meanwhile Jeff is running the entire brain team, and keep pouring contributions on a regular basis: https://research.google.com/pubs/jeff.html https://research.google.com/pubs/jeff.html
- stablemap 9y ago
  There is a paper for the index structure part; several authors. https://arxiv.org/abs/1712.01208 https://arxiv.org/abs/1712.01208
  - snickmy 9y ago
    Ah, great catch! thanks for sharing

codepie 9y ago

Slightly OT, but I wanted to know if there's any group/forum where we can read and discuss papers regularly? I know the morning paper (https://blog.acolyer.org/ https://blog.acolyer.org/), but it is not so interactive, in the sense that you can't have discussions with peers and it turns into just passive reading.

snickmy 9y ago
At Spotify, we have a reading group. I started one at SoundCloud and at Google there were multiple focused on very specific topics. If you can prove to your company the value in investing in a reading group, you can probably start one in your company too. KTH, Stockholm's main technical university, has an open reading group too, open to everyone. Probably if you have a big enough university in your area they have something similar. Alternatively try reddit, ie.: on r/machinelearning. That said, the value of doing this in real life, in front of a whiteboard, is worth going extra mile and find someone to team up with.
almata 9y ago
Not exactly what you're looking for, but maybe you or someone else might be interested on this: http://fermatslibrary.com http://fermatslibrary.com
- shalabhc 9y ago
  Very interesting, thanks! Somewhat related,a general purpose annotation platform is http://hypothes.is http://hypothes.is - a great feature is that PDF annotations are location independent because they use some kind of fingerprint to identify the document.
dilap 9y ago
Yeah, I'd be interested in this too. Something with a lot of discussion, smallish group, to create a sense of social obligation... (So easy to be lazy w/ the infinite world of distractions available.)
AlexCoventry 9y ago
If you're in Boston, we have a reading group for machine learning papers. We'll start up again on the 9th. List of papers we've discussed: https://github.com/pmiller10/cambridge-ai https://github.com/pmiller10/cambridge-ai Mailing list: http://mailman.mit.edu/mailman/listinfo/sipb-deeplearning http://mailman.mit.edu/mailman/listinfo/sipb-deeplearning

yesenadam 9y ago

Could I slip in a request here for contributions to my Ask of 3 days ago, still on top page, but not many answers yet, What are the best scholarly papers you read in 2017? I'm hoping for some discussion on there. https://news.ycombinator.com/item?id=15991373 https://news.ycombinator.com/item?id=15991373

greenpizza13 9y ago

Huge, massive, credibility-undercutting typo in the slot pulling analogy: "you loose." I'm not sure when people started making this mistake, but it's one of my more cringe-worthy ones to see.

thomasfoster96 9y ago
You find a rather common spelling mistake (which has been fixed) “credibility-undercutting” and “cringeworthy”? It seems like you are holding blog posts that have made it to the first page of Hacker News to a ridiculously high standard.

tommyhansen 9y ago

I wish Apple was sharing even more, and not just on Machine learning.